AI Mindset · Model Cheatsheets
Google Gemini

The Google-Native Agent That Can See and Act

Gemini 4 Argon is on sale, and its price doubles on December 1. Google made gemini-4-argon generally available on October 13, 2026 at an introductory $2 per 1M input tokens and $10 per 1M output. From December 1, 2026 it’s $4 and $20. So a pilot costed in October hits a 2x step before the year is out. Second, Google deprecated Gemini 3.7 Flash and 3.5 Flash on October 8 and now silently routes their API calls to 3.8 Flash and 3.6 Flash. Pinned either ID? You aren’t getting the model you pinned. And two shutdown dates moved with no announcement. Re-read the deprecations page before you trust a date on it.

Verified October 15, 2026Gemini 4 Argon GA October 13, 2026 as gemini-4-argonArgon $2 / $10 through November 30, 2026; $4 / $20 from December 13.7 Flash and 3.5 Flash deprecated October 8; calls rerouted to 3.8 and 3.6 Flash2.5 Flash Image shutdown moved to March 15, 2027, unannouncedOmni Flash preview shutdown re-dated to October 22, 2026, unannouncedTen shutdowns dated October 22 to November 17, 2026Gemini 3.1 Pro is still a previewPaid-tier standard; batch is half
1 / Meet Gemini

One intelligence layer across Google’s working surfaces

Gemini isn’t one chat product. It’s a family of models and experiences: Search, the Gemini app, Workspace, Android, developer tools, voice and creative systems. Sold through consumer plans as well as the API.

Personality

Fast, multimodal and connected

Gemini is strongest when the work mixes current information, visual or voice or audio context, and a Google service that can help finish the job.

  • Search and Maps grounding
  • Text, image, voice and video
  • Consumer, enterprise and developer surfaces
Deploy it for

Information-rich action

Use Gemini for current research, Workspace creation, Android assistance, live voice agents, proactive monitoring and agents that need to see and operate interfaces.

  • Research and generative UI
  • Gmail, Docs, Slides and Sheets
  • Browser, mobile and desktop automation
Choose Gemini when

The context already lives with Google

Gemini has a built-in head start when the task begins in Search, Gmail, Drive, Android, Maps, YouTube or a Google developer environment.

  • Less context transfer
  • Live information
  • One model family across many surfaces

Where should Gemini meet the work?

Choose the surface that already owns the context.

Everyday assistant
Gemini app

Use the app for conversation, connected services, multimodal help, Daily Brief and the emerging proactive-agent experiences. Which model you get depends on the AI plan attached to the account. Google says Pro and Ultra subscribers can use Gemini 4 Argon in the app since October 13, 2026.

  • Confirm the in-app model before you standardize on it
  • Voice and camera context
  • Plan tier changes the answer, not just the quota
2 / What’s Current

A frontier model with a price cliff, two Flash models quietly rerouted, and a shutdown calendar that moves

Name the tier or the number means nothing. Every rate quoted here is paid-tier standard, and batch is exactly half. Gemini 4 Argon went generally available on October 13, 2026 as gemini-4-argon. Its introductory $2 and $10 per 1M tokens end on November 30. Gemini 3.8 Flash is still the newest Flash model. Gemini 3.1 Pro is still a preview as of October 15, 2026. On October 8 Google deprecated 3.7 Flash and 3.5 Flash and began routing their calls to newer models. The calendar moved too. The gemini-2.5-flash-image row now reads March 15, 2027, not October 2. The gemini-omni-flash-preview row is back with October 22. And ten rows fall between October 22 and November 17.

$2 / $10
Gemini 4 Argon paid-tier standard per 1M input and output tokens through November 30, 2026
$4 / $20
Gemini 4 Argon paid-tier standard per 1M input and output tokens from December 1, 2026
Oct 13
Gemini 4 Argon general availability as gemini-4-argon; 1M context, March 2026 knowledge cutoff
Oct 8
gemini-3.7-flash and gemini-3.5-flash deprecated; requests routed to 3.8 Flash and 3.6 Flash
Oct 22
Shutdown of three Veo 3.1 previews and the re-dated gemini-omni-flash-preview
Nov 17
Shutdown of three legacy TTS previews and two legacy Live previews
Mar 15, 2027
Shutdown date now shown for gemini-2.5-flash-image, moved from October 2, 2026 with no announcement
May 7, 2027
Earliest shutdown date for gemini-3.1-flash-lite; the pricier gemini-3.5-flash-lite is the named replacement

Gemini 4 Argon GA October 13

Google’s new flagship is buyable, and the clock on its launch price is already running. The ID is gemini-4-argon. Google’s release notes call it “our most capable model built for complex coding, agentic, math, and science tasks,” and the models page lists it as stable. It’s on the paid API through AI Studio for developers and enterprises. Ultra subscribers get it in Antigravity. Pro and Ultra subscribers get it in the Gemini app and AI Mode. The window is 1,048,576 tokens, shared by input, thinking and output. Default output is 65,536 tokens, and long decode takes it to about 1M. The DeepMind model card, published October 13, gives a March 2026 knowledge cutoff. Now the money. $2 per 1M input and $10 per 1M output through November 30, 2026, then $4 and $20 from December 1. Batch runs $1 and $5, then $2 and $10. There’s no free tier. Google’s pitch, from September 30: Argon “delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.” That’s a vendor claim.
· Pricing page, read October 15, 2026: input “$2.00 through November 30, 2026.” then “$4.00 starting December 1, 2026.” The September 30 announcement carries the same $4 and $20 in a footnote.

  • Pin gemini-4-argon and cost it at $4 / $20, not the launch rate
  • The introductory price ends November 30, 2026: 46 days after October 15
  • No free tier, so a free-tier prototype cannot run on it

What breaks when you move a prompt to Argon Breaking changes

Swapping the model ID is the easy part. Google’s migration notes for Argon change four things a working pipeline relies on. Explicit caching “is not supported on Gemini 4 Argon,” so cached input is implicit only, at $0.10 per 1M through November 30 and $0.20 after. temperature, top_p and top_k are ignored. Thinking runs at low, medium or high, with medium as the default. And Google tells you to set max_output_tokens to at least 32,768, and warns that hard prompts can take several minutes. So a pipeline tuned on sampling settings, or one with a 30-second timeout, behaves differently on the same prompt. Test the timeout and the cache bill first. The quality test comes last, right?
· Google’s “What’s new in Gemini 4 Argon” page is a living reference without a dated version, read October 15, 2026. Confirm each setting directly before you ship.

  • Explicit caching is unsupported, so a cache-heavy design needs re-costing
  • Sampling parameters are ignored: steer with thinking level instead
  • Raise output limits and timeouts before the first test run

Gemini 3.8 Flash GA September 2

Still the newest Flash model, and now the landing spot for 3.7 Flash traffic too. Its DeepMind model card, dated September 2, 2026, gave up to 1M tokens of context, 64K tokens of output and a March 2026 knowledge cutoff. Those card figures are unconfirmed as of October 15, 2026. Google’s framing is enthusiastic, and it’s Google’s. The launch post calls it “our most intelligent workhorse model, delivering significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning.” Gemini 4 Argon sits above it, so “the newest Gemini model” and “the newest Flash model” are two different answers.

  • Pin gemini-3.8-flash
  • March 2026 cutoff per the September 2 card, unconfirmed as of October 15, 2026
  • Promotional pricing to December 31, 2026, doubling January 1, 2027
  • Calls to gemini-3.7-flash now land here

The knowledge cutoff Google publishes in one place and omits in the other Card only

Nobody re-checks a negative. That’s how a missing field turns into a false claim. Argon shows the pattern cleanly. The October 13 model card says “The knowledge cutoff for Gemini 4 is March 2026.” The launch blog doesn’t state it. The API model page lists 1,048,576 tokens in and 65,536 out by default, and has no cutoff field at all. Gemini 3.8 Flash looked the same way: its September 2 card gave a March 2026 cutoff, unconfirmed as of October 15, 2026, and its API page has no cutoff field. So a blank field on one Google page isn’t proof that Google hasn’t published the fact. An omission isn’t a denial.

  • Argon’s cutoff, March 2026, is on its October 13 card only
  • The 3.8 Flash cutoff rests on a September 2 card, unconfirmed as of October 15, 2026
  • Where two Google surfaces differ, say which one you trusted

Gemini 3.1 Pro Still a preview

Ask for Pro and a buyer gets gemini-3.1-pro-preview. Note the suffix. Google launched it on February 19, 2026 with the words “we are releasing 3.1 Pro in preview today to validate these updates,” and promised general availability soon. As of October 15, 2026, DeepMind still labels it “Status: Preview,” and the models page lists it only in the all-models table. Google describes it as offering “advanced intelligence, complex problem-solving skills, and powerful agentic and vibe coding capabilities.” Pricing is context-tiered. Paid-tier standard is $2.00 per 1M input and $12.00 per 1M output at 200k tokens or fewer, and $4.00 and $18.00 above 200k. Batch halves both. Look at the Argon launch rate next to that. Same $2 input, cheaper output, and GA. That lasts until November 30.

  • Pin gemini-3.1-pro-preview and price the context tier you will send
  • In preview since February 19, 2026
  • Compare it with Argon at the post-December 1 rate, not the launch rate

Gemini 3.1 Deep Think Documented, not on a plan

DeepMind documents Deep Think. Google’s plans page doesn’t sell it by name. DeepMind’s Gemini 3.1 Deep Think page describes a reasoning mode built on Gemini 3.1 Pro, “Best for modern challenges across science, research and engineering,” with a link to try it in Gemini. But it isn’t on DeepMind’s models index, and the model-card library lists only Gemini 2.5 Deep Think. And the Google AI plans page, read October 15, 2026, doesn’t contain the words Deep Think anywhere. So you can tell a client Google documents a 3.1 Deep Think mode. You can’t tell them which plan includes it.
· DeepMind’s Deep Think page and the plans page are both living pages without a date, read October 15, 2026.

  • A reasoning mode on 3.1 Pro, per DeepMind
  • No model card for 3.1 Deep Think and no plan that names it
  • Confirm access in the product before you quote it

Gemini 3.8 Live and 3.8 Live Extended Thinking GA September 15

Two audio-to-audio Live API models, both generally available and both listed as stable. Google positions gemini-3.8-live as the “default Live API model for most low-latency voice agent experiences without reasoning delays,” and gemini-3.8-live-extended-thinking as the “high-reasoning Live API model for voice interactions.” Both are on the Gemini API and Google AI Studio. The low-latency model is also on Search Live. At launch Google described enterprise access as private preview. Paid-tier standard is $0.75 per 1M input and $4.50 per 1M output for text, with the free tier free of charge. And the old previews now have a date: gemini-3.1-flash-live-preview and gemini-2.5-flash-native-audio-preview-12-2025 shut down on November 17, 2026, with gemini-3.8-live as the replacement.

  • Choose latency or reasoning, then pin the matching ID
  • Legacy Live previews stop November 17, 2026: 33 days after October 15
  • Treat enterprise Live API access as private preview until Google says otherwise

Gemini 3.7 Flash and 3.5 Flash Deprecated October 8, rerouted

A pinned ID is a request, not a guarantee. On October 8, 2026 Google deprecated both models and changed what they return. In its words: “API requests to gemini-3.7-flash are automatically routed to gemini-3.8-flash.” And: “API requests to gemini-3.5-flash are automatically routed to gemini-3.6-flash.” No shutdown date. Neither ID is on the models page any more, and neither has a pricing section. So code that still says 3.7 Flash runs on 3.8 Flash, and your evals describe a model you no longer get. The billing question is open. 3.5 Flash was $1.50 and $9.00 per 1M. Its routing target, 3.6 Flash, is $0.75 and $3.75 on promotion. No Google page says which rate a routed call is billed at.
· Both routing statements are from Google’s API release notes, which are rewritten in place, and the 3.7 Flash model page banner, read October 15, 2026.

  • Re-pin to gemini-3.8-flash or gemini-3.6-flash explicitly, so the ID in code matches the model you get
  • Re-run your evals on the routed model
  • Ask Google in writing which rate a routed 3.5 Flash call is billed at

The recycled superlative Recycled claim

Three Flash generations. One adjective. At its August 13 launch Google called Gemini 3.7 Flash “our most intelligent workhorse model yet for coding and agents.” Three weeks later near-identical framing belonged to 3.8 Flash. Five weeks after that, 3.7 Flash was deprecated. The words were accurate on each day and distinguishing on none. So treat a vendor superlative as a marketing constant. Take the differences from the model card, the price table and your own traces.

  • The same superlative has covered three Flash generations
  • Quote it with its date attached or not at all
  • Benchmark the models yourself rather than reading the adjectives

Gemini 3.6 Flash Two generations back

The July 21 default, still listed, still on the same promotional pricing as 3.8 Flash, and now the model behind every gemini-3.5-flash call. The models page describes it as “Our previous-generation Flash model, balancing speed and multimodal capabilities across general agentic and everyday tasks.” Its distinguishing claim travels with it. Google wrote that “according to the Artificial Analysis Index, it reduces output token usage by 17% compared to 3.5 Flash.” One gap to know. It isn’t on the computer-use model list.

  • Pin gemini-3.6-flash if you are staying on it
  • The 17 percent figure is the Artificial Analysis Index, cited by Google
  • Not listed for computer use
  • Same promotional rate, same January 1 step-up

Gemini 3.1 Flash-Lite Shuts down May 7, 2027

Google has put a shutdown date on the cheapest model it sells. The ID is gemini-3.1-flash-lite, and it runs at $0.25 per 1M input tokens for text, image and video and $1.50 per 1M output. The deprecations page schedules it to stop on May 7, 2027. The named replacement is gemini-3.5-flash-lite, at $0.30 and $2.50. That’s 20 percent more on input and 67 percent more on output. Meanwhile the models page still sells 3.1 Flash-Lite as “Frontier-class performance rivaling larger models at a fraction of the cost,” with no deprecation mark. So the catalog and the schedule tell a buyer two different things about the same model.

  • Any high-volume pipeline pinned here has a migration with a price increase attached
  • Google publishes May 7, 2027 as the earliest possible date: a planning floor
  • Price the replacement now: 204 days separate October 15 from the published date

Gemini 3.5 Flash-Lite The named replacement

The pricier Flash-Lite is the one with a future. List price is $0.30 per 1M input tokens and $2.50 per 1M output on the paid standard tier. Google reports roughly 350 output tokens per second. It’s the replacement Google names for 3.1 Flash-Lite, it carries no shutdown date of its own, and it’s on the computer-use model list. Google’s prompt injection detection for computer use applies to “Gemini 3.5 Flash-Lite or later,” which makes it the cheapest model with that control.

  • Pin gemini-3.5-flash-lite for new high-volume work
  • Budget the step up from $0.25 / $1.50 if you are moving off 3.1 Flash-Lite
  • The cheapest model that gets prompt injection detection

Gemini 3.5 Pro Still “coming soon”

A banner isn’t a date. DeepMind’s Gemini 3.1 Pro page carries a line reading “3.5 Pro coming soon” above the 3.1 Pro heading, with a link to learn about Gemini 3.5. That’s the whole promise. On May 19, 2026 Google said Pro was “already being used internally, and we look forward to rolling it out next month.” The month it named was June. On July 21 Google said only that it’s testing with partners. It’s October. The models page and the pricing page have no 3.5 Pro, and the Argon launch posts don’t mention it. So say what Google says, coming soon, and attach the date it first said next month.
· DeepMind’s Pro page is a living page without a date, read October 15, 2026.

  • The month Google named was June; it is now October
  • Quote “coming soon” with the May 19 promise beside it, never a date
  • Gemini 3.1 Pro is the Pro model that exists, and Argon sits above both

Computer use Argon listed, in preview

The computer-use page names four models: gemini-4-argon, gemini-3.8-flash, gemini-3.5-flash-lite and gemini-3-flash-preview. Google describes Argon as “Our most capable model, supporting computer use for complex reasoning and agentic UI workflows,” while Argon’s own model page marks computer use “Supported (Preview).” gemini-3.6-flash isn’t on the list. The page covers three environments, browser, mobile and desktop, plus thinking-level control, configurable safety policies and safety overrides. Now the control that matters. Prompt injection detection applies to Gemini 3.5 Flash-Lite or later, and it’s off by default. That’s a real control. Turn it on.
· Computer use page, stamped October 13, 2026 and read October 15, 2026.

  • Pin a listed model; 3.6 Flash is not one
  • Argon computer use is preview, so scope it as a pilot
  • Prompt injection detection defaults to off: switch it on in config

Gemini 3.8 Flash Cyber Fairwind only

The cyber model moved up three generations, from Gemini 3.5 Flash Cyber to Gemini 3.8 Flash Cyber, and changed its access route on September 2, 2026. It’s “only available to trusted defenders who require a more comprehensive set of cyber capabilities.” Access runs through the Fairwind Program, “a limited access program for governments and trusted partners to use our cyber defense tools.” Google says “CodeMender with Gemini 3.8 Flash Cyber delivers the specialized reasoning to write and validate code fixes, at a fraction of the operating cost.” On October 13 Google said cyber defenders can apply for Argon without cyber guardrails on a rolling basis, and claimed Argon solved all 12 of its simulated cyberattack scenarios, up from five. That’s Google grading Google.

  • Not a purchasable capability
  • The fate of Gemini 3.5 Flash Cyber is unconfirmed rather than concluded
  • CodeMender is available more widely than the cyber model is

Two text-to-speech models, two prices Stable September 22

Voice output stopped being a preview, and the two models don’t cost the same. The first, gemini-3.8-flash-tts, is the “flagship creative text-to-speech model for studio-grade voice fidelity, expressive acting, Voice design, and Voice replication.” The second, gemini-3.8-flash-lite-tts, is the “fast, cost-efficient text-to-speech model for high-volume production, real-time voice agent cascades, and Voice replication.” Flash TTS is $0.50 per 1M input and $9.00 per 1M audio output through December 31, 2026, then $1.00 and $18.00. Flash-Lite TTS output is $6.00 through December 31, then $12.00. Both double on the day the Flash text promotion ends. And three old preview IDs, gemini-3.1-flash-tts-preview, gemini-2.5-flash-preview-tts and gemini-2.5-pro-preview-tts, shut down on November 17, 2026.
· The September 22, 2026 GA date comes from Google’s API release notes, which Google rewrites in place.

  • Pin the explicit TTS ID, and say which one in the contract
  • Price the 2027 rate for anything running past December 31
  • Voice replication needs written consent from the person whose voice it is
Model or capabilityPaid-tier standard, per 1M tokensWhat a buyer needs to know
gemini-4-argon$2.00 in / $10.00 out through November 30, 2026; $4.00 / $20.00 from December 1, 2026Batch $1.00 / $5.00, then $2.00 / $10.00. Cached input $0.10, then $0.20, implicit only. No free tier.
gemini-3.8-flash, gemini-3.6-flash$0.75 in / $3.75 out through December 31, 2026, then $1.50 / $7.50Batch $0.375 / $1.875, then $0.75 / $3.75 from January 1. Free tier is free of charge. These are the routing targets for 3.7 Flash and 3.5 Flash.
gemini-3.7-flash, gemini-3.5-flashNo pricing sectionDeprecated October 8, 2026. Calls route to 3.8 Flash and 3.6 Flash. Google does not say which rate a routed call is billed at.
gemini-3.1-pro-preview$2.00 in / $12.00 out at 200k tokens or fewer; $4.00 in / $18.00 out above 200kContext-tiered. Batch halves both tiers: $1.00 / $6.00 and $2.00 / $9.00.
gemini-3.1-flash-lite$0.25 in / $1.50 outCheapest on both sides, and scheduled to shut down on May 7, 2027. Input covers text, image and video.
gemini-3.5-flash-lite$0.30 in / $2.50 outThe named replacement for 3.1 Flash-Lite: 20 percent more on input, 67 percent more on output.
gemini-3.8-flash-tts$0.50 in / $9.00 audio out through December 31, 2026, then $1.00 / $18.00Batch $0.25 / $4.50, then $0.50 / $9.00.
gemini-3.8-flash-lite-tts$6.00 audio out through December 31, 2026, then $12.00Batch output $3.00, then $6.00. Input rate unconfirmed.
gemini-3.8-live and gemini-3.8-live-extended-thinking$0.75 in / $4.50 out for textFree tier free of charge.
gemini-omni-1.1-flash$1.50 in for text, image, video and audio / $9.00 out for textNo free tier. Named replacement for the Omni preview and three Veo 3.1 previews, all stopping October 22, 2026.
gemini-3.5-transcribe$2.00 or $0.003 per audio minute in / $12.00 or $0.002 per minute outBatch transcription is the cheaper of the two IDs by a wide margin.
gemini-3.5-transcribe-live$3.50 or $0.005 per audio minute in / $21.00 or $0.004 per minute out75 percent more per 1M tokens than the batch model; 67 percent more per minute on input, 100 percent more on output.
gemini-nano-banana-2.1 (Nano Banana 2.1)$1.50 in / $7.50 out for text and thinking plus $30.00 out for imagesGA October 6, 2026. Batch $0.75 / $3.75 / $15.00. No free tier. Named replacement for 3.1 Flash Image.
gemini-3.1-flash-image (Nano Banana 2)$0.50 in / $3.00 out for text and thinking plus $60.00 out for imagesDeprecated October 6, 2026 with no shutdown date. Google states the image rate as $0.045 per 0.5K image. No free tier.
gemini-3.1-flash-lite-image (Nano Banana 2 Lite)$0.25 in / $30.00 out for imagesNow the named replacement for gemini-2.5-flash-image.
gemini-3-pro-image (Nano Banana Pro)$2.00 in / $12.00 out for text and thinking plus $120.00 out for imagesBatch $1.00 in / $6.00 text out. Priority $3.60 / $21.60 / $216.00. The most expensive image output Google lists.
Search grounding5,000 free requests per month, then $14 per 1,000Shared across all Gemini 3 and newer models, Argon included.
Shutdown dateModelReplacement named by Google
June 25, 2026 (as the row now reads)gemini-3.1-flash-image-previewgemini-nano-banana-2.1. The page named this same preview ID as a live replacement for an October 2 shutdown, and the ID appears nowhere on the models page
August 31, 2026 (passed; described as deprecated, never as shut down)gemini-robotics-er-1.6-previewgemini-robotics-er-2-preview. The models page still lists the 1.6 preview with no mark
October 5, 2026 (passed; no outcome stated)antigravity-preview-05-2026antigravity-preview-09-2026, which uses a different tool contract. The models page lists only the September agent
October 22, 2026veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-previewgemini-omni-1.1-flash. The models page still lists Veo 3.1 previews
October 22, 2026 (re-dated from September 30, unannounced)gemini-omni-flash-previewgemini-omni-1.1-flash. The pricing page still sells the preview
October 23, 2026deep-research-pro-preview-12-2025deep-research-preview-04-2026 or deep-research-max-preview-04-2026
November 17, 2026gemini-3.1-flash-tts-preview, gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-ttsgemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
November 17, 2026gemini-3.1-flash-live-preview, gemini-2.5-flash-native-audio-preview-12-2025gemini-3.8-live
March 15, 2027 (moved from October 2, 2026, unannounced)gemini-2.5-flash-imagegemini-3.1-flash-lite-image, changed from gemini-3.1-flash-image-preview. The models page lists the model with no deprecation mark
May 7, 2027gemini-3.1-flash-litegemini-3.5-flash-lite, which costs more on both input and output
No date (deprecated October 6, 2026)gemini-3.1-flash-imagegemini-nano-banana-2.1. The models page still lists 3.1 Flash Image as stable
No date (deprecated October 8, 2026; calls already routed)gemini-3.7-flash, gemini-3.5-flashgemini-3.8-flash and gemini-3.6-flash, which already answer requests sent to the old IDs

How much action should Gemini take?

Move from information to action only as permissions and consequence allow.

ObserveAct
Prepare
Draft the next step

Gemini prepares the email, plan, document, route or interface. It doesn’t change anything outside.

  • Useful default for work
  • Easy human review
  • Keep assumptions visible
3 / The Stack

Gemini’s advantage is the stack around the model

Each surface brings a different kind of context. Search knows the live web. Workspace knows the work. Android knows the moment. And the API turns the model into a building block. Three changes to act on. Gemini 4 Argon reached the app and AI Mode for Pro and Ultra subscribers on October 13. Skills replaced Gems, and the Gems retirement has dates, while the Workspace skills rollout that should replace it is paused. And Gemini 3.8 Live with Live Avatar shipped into Gemini Enterprise. Meanwhile the consumer Gemini Drop has missed two months outright. This is also where a buyer meets the consumer plans, which decide which model an account gets, and how much of it.

Assistant

Gemini app

Conversation, connected apps, multimodal input, Daily Brief and emerging personal agents. Google lists a Windows app in its September 2026 roundup. What the account can reach depends on its AI plan, and Argon arrived for Pro and Ultra subscribers on October 13, 2026.

Information

Search AI Mode

Live web grounding, information agents, Maps and custom generative interfaces. Argon reached AI Mode for Pro and Ultra subscribers on October 13, 2026.

Work

Google Workspace

Gmail, Docs, Slides, Sheets, Meet and organizational content, with Gemini 3.8 Live Extended Thinking rolling out to Workspace and Gemini Live subscribers in Docs, Gmail and Keep.

Research

Gemini Notebook

Source-grounded notebooks, audio and video overviews, reports and study tools. Renamed from NotebookLM on July 16, 2026, with a secure cloud computer for code execution and sync with the Gemini app.

Development

Gemini API

Model access, built-in tools, computer use, embeddings and application-specific agents. Argon is on the paid tier only.

Coding

Antigravity

Agent-first development with artifacts, plans, screenshots and execution evidence. The default agent changed on September 17, 2026, and Argon reached Ultra subscribers here on October 13.

Device

Android

Camera, voice, notifications, connected apps and contextual assistance close to the moment of work. At Galaxy Unpacked on July 22, 2026 Google said Gemini Intelligence task automation reaches more than 40 apps. Gemini Notebook comes preinstalled on the Galaxy Z Fold8, and Gemini runs on the Galaxy Watch 9.

Creative

Omni, Veo, Lyria and image tools

Image, video and music creation connected to Gemini reasoning and the wider Google ecosystem. Three Veo 3.1 previews stop on October 22, 2026.

September 30 to November 17

Skills replaces Gems, and the Workspace half is paused

A feature your users built things in is going away, and its replacement is late in exactly the place you manage. Skills are invoked by typing a forward slash and the skill name, they stack, and they take reference files. Here’s the calendar. Google says it will “remove support for Gems starting in November” for personal accounts. On November 17, 2026 Gems move into the Settings panel. Workspace retirement is “No sooner than March 1, 2027” for business, enterprise and nonprofit, and “No sooner than June 1, 2027” for education. Those are floors, not dates. Google says it will convert existing Gems to skills automatically. But on October 13 it paused skills in Workspace apps, “Delayed until further notice,” while the Gemini app rollout continues to mid-November. So an admin is holding a retirement date with no replacement in the apps yet. Inventory who has built a Gem before November 17.

September 24

Gemini 3.8 Live with Live Avatar, in Gemini Enterprise

Google’s words: “Starting today, Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise.” Real-time animated avatars with lip-sync across 97 languages, asynchronous tool execution so the avatar keeps talking while a tool runs, and SynthID watermarking on the output. This is the first Live capability to land inside an enterprise product rather than a private preview. Avatar is a Gemini Enterprise feature. Direct enterprise access to the Live API models was private preview at launch. Those are two different purchases.

September 17

A new Antigravity agent, and the old one’s date passed in silence

Google’s API release notes record antigravity-preview-09-2026 replacing the May 2026 version. Built-in tools moved to PascalCase parameters, and file edits became line-range replacements instead of full rewrites. Both are breaking changes for any integration written against the old contract. The old agent, antigravity-preview-05-2026, was due to shut down on October 5, 2026. As of October 15, 2026, the deprecations page still shows that row, the models page lists only the September agent, and no Google page says the old one stopped. Eighteen days from release to shutdown wasn’t a migration window. If anything you own still calls the old ID, find out by testing it, not by reading.

September 15

Gemini 3.8 Live reached GA

Two audio-to-audio models, gemini-3.8-live for low latency and gemini-3.8-live-extended-thinking for reasoning, generally available through the Gemini API and Google AI Studio. The low-latency model is also on Search Live. The same release marked gemini-3.1-flash-live-preview as legacy. It now has a shutdown date: November 17, 2026.

September 3

Lyria 3.5 reached GA

A full-length song generation model with, in Google’s words, “improved musical coherence, natural vocals, and fine-grained duration and structural control,” generating 44.1 kHz stereo audio from text and image inputs. The models page lists lyria-3.5 as “Our flagship music generation model, optimized for full-length songs with complex structural coherence,” and the deprecations page names it as the replacement for lyria-3-pro-preview, with no shutdown date yet. DeepMind’s model-card index dates the Lyria 3.5 card July 29, 2026. The September 3 GA date comes from Google’s API release notes.

September 1

Agentic video understanding

Google introduced agentic video understanding for Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. Google claims up to 88% fewer tokens, 66% lower cost and 7% better accuracy, at no extra charge. Those are Google’s figures. Here’s the catch. 3.7 Flash calls now route to 3.8 Flash, a model the announcement never named. So a pipeline built on the 3.7 Flash numbers is running on a model nobody measured for this. Test on your own footage before you rebuild around the savings.

August 27

Gemini Omni Flash

The model ID is gemini-omni-1.1-flash. It adds video extension, interpolation and configurable resolution at 360p, 720p, 1080p and 4k, at $1.50 per 1M input and $9.00 per 1M output with no free tier. Google’s pages give different answers about its status. The models page lists it under a preview heading. DeepMind and the release notes say GA. The dated announcement calls it “production-ready.” It’s now the named replacement for the Omni preview and three Veo 3.1 previews, all dated October 22, 2026.

August 26

Gemini 3.5 Transcribe

Two model IDs, gemini-3.5-transcribe and gemini-3.5-transcribe-live, with support for 85 or more languages, speaker diarization and custom vocabulary biasing. The live model costs 75 percent more per 1M tokens than the batch model, and between 67 and 100 percent more per audio minute depending on direction. So the choice between them is a budget decision as well as a latency one.

Embeddings

Gemini Embedding 2

Google calls Gemini Embedding 2 its “first multimodal embedding model, mapping text, images, video, audio, and PDFs into a unified embedding space.” Check the ID before you build on it. The models page lists gemini-embedding-2-preview. The deprecations page lists gemini-embedding-2 with no shutdown date, and an embedding-2-preview that shut down on August 10. A unified space across five modalities is a real retrieval capability. An index you build on the wrong ID is a re-embedding bill later.

October 1

The September Workspace Feature Drop

Google Pics, a canvas in Sheets, cross-app tasks, Ask Gemini in Chat, video summaries in Vids, and Gemini-based data classification in Drive. That last one is the admin item. It feeds data loss prevention and retention labels, which means a model is now deciding how your files get classified. Review it before you enable it. Workspace features arrive by edition and admin policy, so confirm eligibility before you describe any of them as available.

Cadence break

Two consecutive months with no Gemini Drop

The July 31, 2026 drop is the most recent edition Google has published on its blog. No August or September edition exists. And Google’s own Drops hub lists June 2026 as its latest drop and doesn’t list July at all. So the channel Google built to announce drops is behind the drops themselves. October isn’t due yet: the July edition published on the 31st, so the test is October 31. Two missed months is a gap. Three would be a pattern.

August 12

New connected apps and services

Google added Granola, Otter.ai, Wix, OpenTable (UK), Ticketmaster, Zocdoc and others to the Gemini app’s connected apps on August 12, 2026. Every connector widens what the assistant may read and act on. So treat the list as a permissions decision, not a feature announcement.

SurfaceContext it ownsBest outcomeAvailability caution
Gemini appPersonal conversation and connected appsEveryday help and proactive assistanceFeatures vary by plan, language and region; Argon is named for Pro and Ultra as of October 13, 2026
Search AI ModeLive web, shopping and MapsGrounded answer or interactive interfaceGenerative UI and agents roll out over time
WorkspaceOrganizational mail, files and meetingsNative work artifactEdition and admin policy matter; skills in Workspace apps delayed until further notice as of October 13, 2026; Gems retire no sooner than March 1, 2027 (business, enterprise, nonprofit) and June 1, 2027 (education)
Voice and LiveSpoken conversation in real timeProduction voice agents and hands-free helpGA through the API and AI Studio since September 15, 2026; Live Avatar in Gemini Enterprise since September 24, 2026; legacy Live previews stop November 17, 2026
Gemini NotebookCurated user source setSource-grounded learning and briefingThe cloud computer is on AI Ultra and Workspace Expanded Access; the July 16, 2026 post promised Pro on the web in coming weeks
API / AntigravityDeveloper-defined tools and environmentCustom agent or software workflowRequires safety and permission design; Argon is paid tier only, and in Antigravity for Ultra subscribers
AndroidDevice, camera and contextual momentPersonal assistance and actionHardware and rollout vary
PlanWhat Google lists in it, October 15, 2026Price, with the date Google stated it
Google AI Plus400 GB storage, access to Gemini 3.6 Flash and Deep Research; the card is labeled “Includes Gemini 4 Argon” while its bullets do not list Argon$7.99 a month in the US, stated January 27, 2026; price varies by country, and the plans page shows no amount
Google AI Pro5 TB storage, access to Gemini 4 Argon, Google Flow, $10 a month in Google Cloud credits$19.99 a month, stated August 19, 2026 in the student-offer post
Google AI Ultra 5x5x the Pro plan, 20 TB storage, higher Argon access, $40 a month in Google Cloud credits$200 or $100 a month; Google has not said which tier is which. Both stated May 19, 2026
Google AI Ultra 20x20x the Pro plan, 30 TB storage, higher Argon access, $100 a month in Google Cloud credits$200 or $100 a month; Google has not said which tier is which. Both stated May 19, 2026
4 / Agentic Work

Google is turning Search and the device into persistent workers

The Gemini 3.5 through 4 line is built to sustain work, not just answer prompts. Spark, background information agents, voice agents and computer use are Google trying to make intelligence persistent across time and interfaces. Most of those surfaces are still limited, regionally excluded or in preview. The control worth having is first-party prompt injection detection for computer use. It ships off by default, so somebody has to decide to turn it on.

A safe computer-use workflow

Computer use expands capability and the attack surface at the same time.

Define
Constrain the environment

Name the application, the account, the data, the actions and the stopping conditions that belong to this task. Pick a model on the computer-use list, because Gemini 3.6 Flash is not one.

  • Least-privilege access
  • No implicit cross-account reach
  • Explicit destructive-action rule

Gemini Spark Not in the EEA, Nigeria, Switzerland or the UK

A personal agent meant to help across the digital day. The July Gemini Drop framed the release as “Gemini Spark is going global... now available worldwide.” Google’s support page states the real list: “Available wherever Gemini Apps are supported, except in the European Economic Area, Nigeria, Switzerland and the United Kingdom.” So read the exclusion list, not the headline. And check the plan requirement in the product, because the plans page no longer names Spark.

  • “Worldwide” with four named exclusions, unchanged since July
  • Unavailable to readers in the UK, the EEA, Switzerland and Nigeria
  • Plan gating is no longer published on the plans page

Computer use Detection ships off

The most useful thing in the computer-use docs is a defense, and it’s switched off until you switch it on. Google documents prompt injection detection for Gemini 3.5 Flash-Lite or later, alongside configurable safety policies, safety overrides and thinking-level control, across browser, mobile and desktop. The default is false. The page lists four supported models: gemini-4-argon, gemini-3.8-flash, gemini-3.5-flash-lite and gemini-3-flash-preview. Argon’s support is marked preview. The capability launched on Gemini 3.5 Flash on June 24, 2026. That model now routes to 3.6 Flash, which isn’t on the list.

  • Set prompt injection detection to on in every deployment
  • Browser, mobile and desktop are separate environments with separate risk
  • Code still pinned to gemini-3.5-flash now lands on an unlisted model

Managed agents and the Antigravity default Old agent’s date passed October 5

A model swap shows up in a dashboard. A tool-contract change shows up as failing agents. Google opened managed agents to free-tier projects on July 28, 2026: “Developers can experiment with agentic workflows using an API key from a project without active billing.” That post named antigravity-preview-05-2026 as running Gemini 3.6 Flash by default. Google’s API release notes, dated September 17, 2026, record antigravity-preview-09-2026 replacing it, with PascalCase tool parameters and line-range file edits. Both are breaking changes. The deprecations page dated the May agent’s shutdown October 5, 2026. That date passed with no outcome stated anywhere.

  • Treat antigravity-preview-05-2026 as gone and test anything that still calls it
  • PascalCase tool parameters and line-range file edits are breaking changes
  • Argon reached Ultra subscribers in Antigravity on October 13, 2026

Watching the agent while it runs Private preview

Two posts on Google’s developer blog give the safe workflow above a control plane. On September 16, 2026 Google described Agent Anomaly Detection, in private preview on the Gemini Enterprise Agent Platform. It’s an oversight layer that analyzes OpenTelemetry traces to catch behavioral risks without affecting live request performance. On September 15 Google published guidance on building zero-trust AI agents that judge intent rather than syntax. Read both before an agent touches production. Neither was generally available when announced, which makes prompt injection detection the control you can switch on now.

  • Trace-based oversight rather than prompt-level filtering
  • Private preview at announcement, so do not scope a program around it
  • Prompt injection detection is the shipped alternative

Voice agents Avatar shipped September 24

Gemini 3.8 Live and 3.8 Live Extended Thinking make production voice agents a supported build rather than a preview experiment. The low-latency model is aimed at conversational turn-taking. The extended-thinking model is for problems that need reasoning mid-call. And on September 24, 2026 Google put a face on it: “Starting today, Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise.” That means lip-synced animated avatars across 97 languages, asynchronous tool execution and SynthID watermarking. Any voice agent still on a legacy Live preview has until November 17, 2026.

  • Latency and reasoning are separate model choices
  • Live Avatar is a Gemini Enterprise feature, not a Live API entitlement
  • Design the escalation path to a human before you launch

Agentic video understanding September 1

Google’s agentic video capability was announced for Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, at no extra cost. Google claims up to 88% fewer tokens, 66% lower cost and 7% better accuracy. One of those three models is now deprecated, and its calls go to 3.8 Flash, which the announcement never named.

  • Vendor figures, attributed to Google
  • Named for 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite
  • Validate the savings on your own video, on the model that answers

Information agents in Search

Agents can watch an ongoing information need and report what changed, with links for deeper action.

  • Repeated research
  • Alerts and monitoring
  • Source review remains necessary

Daily Brief

The Gemini app can pull timely personal information into one proactive morning view.

  • Reduce manual checking
  • Depends on connected context
  • Review privacy and relevance

Generative interfaces

Search can build a purpose-built interface or simulation instead of forcing every answer into prose. Google promised this for everyone in Search “this summer” on May 19, 2026. The summer it named ended on September 22. Treat it as promised until a dated Google post says it shipped, and check in Search before you describe it to a client.

  • Promised for a window that has closed
  • Interactive decision support
  • Generated logic still needs testing

Android Halo

A status-bar surface that shows agent activity, announced May 19, 2026 in preview and described as rolling out later this year. It’s a visibility surface, not a management console. Check the device before you promise it.

  • Agent activity, not agent control
  • Announced in preview
  • Rollout status matters
5 / Creation

Gemini is joining reasoning and media production

Image output is charged at rates two orders of magnitude above text, and the image models have no free tier at all. Nano Banana Pro now has a published price, and it’s the steepest: $120 per 1M image output tokens. Start there. What the creative stack is good for is connecting a source-rich research process straight to image, video, music and presentation outputs, without shrinking creativity down to a single model name.

Nano Banana family Four IDs, one deprecated

Google’s image stack is four models with very different economics, all paid, none with a free tier. Image output has two parts, text and thinking plus images, and the first one gets missed. Nano Banana 2.1 is gemini-nano-banana-2.1, GA October 6, 2026, at $1.50 in, $7.50 text out and $30.00 image out. It’s the named replacement for Nano Banana 2, gemini-3.1-flash-image, which Google deprecated the same day with no shutdown date. That older model is $0.50 in, $3.00 text out and $60.00 image out. So the upgrade costs more on input and text and less on images. Nano Banana 2 Lite, gemini-3.1-flash-lite-image, is $0.25 and $30.00. Nano Banana Pro, gemini-3-pro-image, is $2.00 in, $12.00 text out and $120.00 image out, with priority at $216.00. Re-cost before you migrate, because which model is cheaper depends on how many images you make per prompt.
· Nano Banana 2.1’s DeepMind model card, published October 6, 2026, lists the Gemini app, AI Studio, the Gemini API, AI Mode, Google Ads, Flow and Stitch.

  • Pin gemini-nano-banana-2.1 for new image work
  • Image output has a text-and-thinking component as well as a per-image one
  • Nano Banana Pro is the most expensive image output Google lists

Gemini Omni Flash GA claimed, preview listed

The multimodal creation model is gemini-omni-1.1-flash. It adds video extension, interpolation and configurable resolution at 360p, 720p, 1080p and 4k, at $1.50 per 1M input and $9.00 per 1M output with no free tier. Its status depends which Google page you read. DeepMind and the release notes say GA. The API models page lists it under a preview heading. The August 27, 2026 announcement says production-ready. The preview it succeeds, gemini-omni-flash-preview, was due to stop September 30. Its row vanished and came back dated October 22, 2026, and the pricing page still sells it.

  • Pin gemini-omni-1.1-flash, not the preview
  • Say “Google describes it as production-ready,” not “it is GA”
  • Move anything on the preview before October 22, 2026

Text to speech Stable September 22

Two voice models went stable, and three old ones now have a shutdown date. The flagship, gemini-3.8-flash-tts, is built for studio-grade fidelity, expressive acting, voice design and voice replication. The cheap, fast one for high-volume production is gemini-3.8-flash-lite-tts. Flash TTS is $0.50 in and $9.00 audio out through December 31, 2026, then $1.00 and $18.00. Flash-Lite TTS output is $6.00, then $12.00. The three old previews, gemini-3.1-flash-tts-preview, gemini-2.5-flash-preview-tts and gemini-2.5-pro-preview-tts, all stop on November 17, 2026. Voice replication is the feature to govern, not the price. Get written consent from the person whose voice you clone, and keep it.

  • Three legacy TTS previews stop November 17, 2026
  • Both output rates double on January 1, 2027
  • The GA date comes from Google’s release notes, which are rewritten in place

Lyria 3.5 GA September 3

Google’s full-length song generation model, with a pinnable ID: lyria-3.5. The models page calls it “Our flagship music generation model, optimized for full-length songs with complex structural coherence.” Google describes improved musical coherence, natural vocals and fine-grained duration and structural control, generating 44.1 kHz stereo audio from text and image inputs. The deprecations page names it as the replacement for lyria-3-pro-preview, which has no shutdown date yet. Music generation is listed inside the consumer AI plans.

  • Pin lyria-3.5
  • Plan the move off lyria-3-pro-preview before it gets a date
  • Rights clearance matters more for music than for most generated media

Gemini 3.5 Transcribe GA August 26

Speech-to-text in two forms, gemini-3.5-transcribe and gemini-3.5-transcribe-live, covering 85 or more languages with speaker diarization and custom vocabulary biasing. Vocabulary biasing is the feature that makes internal product names and people’s names survive a transcript. The live model is 75 percent more expensive per 1M tokens than the batch model. Per minute, input is $0.005 against $0.003, which is 67 percent more, and output is $0.004 against $0.002, which is 100 percent more.

  • Separate IDs for batch and live, with a real price gap
  • Diarization for multi-speaker recordings
  • Bias the vocabulary before you judge accuracy

Veo and Flow Veo 3.1 previews stop October 22

Google’s video stack supports cinematic clips and scene-oriented workflows, not just a single generated shot. Three Veo IDs have a near date: veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview shut down on October 22, 2026, and Google names gemini-omni-1.1-flash as the replacement. The models page still lists Veo 3.1 previews. That date appears only on the deprecations page. Google Flow is listed inside the Google AI Pro plan and above.

  • Move Veo 3.1 preview calls to Omni 1.1 Flash before October 22
  • Use references and continuity
  • Plan for post-production

Slides and Workspace

Gemini can help move research and narrative into presentation form inside the same organizational environment. The September Workspace Feature Drop, published October 1, 2026, adds Google Pics, a canvas in Sheets, cross-app tasks and video summaries in Vids.

  • Use brand templates
  • Link claims to sources
  • Confirm edition eligibility for each drop feature

Gemini Notebook briefing outputs

Audio, video, reports, quizzes and other outputs turn a curated source set into several learning formats. Google reports more than 30 million users and over 600,000 organizations on the product.

  • Source-grounded transformation
  • Useful for staff training
  • Keep audience needs explicit

Google Vids July 16

Vids gained Gemini Omni clip generation, and personal avatars built from a selfie and a voice sample, watermarked with SynthID. Google lists it for Google AI Pro and Ultra, plus Workspace business editions.

  • Check plan eligibility first
  • Get written consent for any likeness
  • Disclose synthetic presenters

Generative UI

Custom charts, tools and simulations are a new kind of creative output for questions that need interaction. Google promised general Search availability “this summer” on May 19. Treat it as promised, not shipped.

  • Explain complex systems
  • Let users explore variables
  • Test the generated behavior
Multimodal campaign brief
Using the research notebook and brand system, create three campaign territories. For each, provide the strategic idea, hero image direction, six-second video beat and the evidence that makes the concept credible.
Source-grounded presentation
Turn this Gemini Notebook source set into an executive presentation. Keep one claim per slide, cite the exact source behind each claim and use our approved Slides theme.
Generative decision tool
Build an interactive comparison tool for these options using current Search and Maps data. Let the user change budget, distance and priority, and show how the recommendation changes.
Creative truth review
Inspect every generated visual for inaccurate text, impossible product details, misleading data or implied claims that the source material does not support.
6 / Behavior Playbook

Ten habits for working across Google’s surfaces

Pick the surface that already owns the context. Pin what you build on, then check it still answers as itself. Cost the tier, and the date, the work will actually run on. And keep the evidence attached as the work moves toward action. That’s the whole behavior, right?

01

Start in the context owner

Use Search, Workspace, Gemini Notebook, Android or the API, depending on where the evidence already lives.

02

Ask for grounding

Use live Search and Maps when current external reality matters. The free grounding allowance is shared across every Gemini 3 and newer model, so budget it once.

03

Curate when authority matters

Use Gemini Notebook or a controlled source set instead of the whole web.

04

Use the actual modality

Give Gemini the image, the video, the voice or the screen state instead of describing it badly. Voice is a first-class production build now, not a demo.

05

Separate prepare from act

Let Gemini draft first. Require approval before anything consequential happens outside.

06

Name the surface and the plan

Availability differs across app, Search, Workspace, API, region and subscription, and the consumer plan decides whether an account reaches Argon at all.

07

Cost the tier and the date

Say whether a figure is free tier, paid standard or batch, and when it changes. Argon doubles on December 1, 2026. The Flash and TTS promotions end December 31.

08

Preserve source links

Keep live information attached as it moves into a document, deck or decision.

09

Discount the superlatives

Google has called three Flash generations its most intelligent workhorse model, and deprecated one of them eight weeks after its own launch. Compare model cards, prices and your own traces instead.

10

Pin the model, then watch it

Name gemini-4-argon, gemini-3.8-flash or another explicit ID. Then read the release notes on a schedule, because Google can route a pinned ID to a different model without a shutdown date.

7 / Watch Outs

An ecosystem this broad is easy to overstate

Three mistakes, in the order people make them. Turning a limited preview into a universal capability statement. Trusting a documentation page to describe the present. And trusting a pinned model ID to stay the model you tested. Gemini features arrive across products, plans and regions at different times, and the deprecation page moves dates without saying so. That’s what makes all three so easy.

A pinned ID can be rerouted

You can do everything right and still not get the model you asked for. On October 8, 2026 Google began routing gemini-3.7-flash calls to gemini-3.8-flash and gemini-3.5-flash calls to gemini-3.6-flash. No shutdown date. No error. No billing note. Your code keeps working, and your evals quietly describe something else. So re-pin to the model that actually answers, and re-run the tests that justified the choice.

An absence is a claim too

Nobody re-checks a negative. That’s what makes it the dangerous kind. Google fills blanks without announcing it. Nano Banana Pro sat on the models page with no pricing section; it now has one, at $120 per 1M image output tokens. Argon’s knowledge cutoff appears on its model card and nowhere else. A statement that something doesn’t exist expires exactly like a statement that it does. So check the model card, the pricing page and the API reference before relying on a claim that something is missing.

The vendor can contradict itself

The Plus card says “Includes Gemini 4 Argon” and its bullets don’t. The app post gives Argon to Pro subscribers, and the pricing page says Argon isn’t on the AI Pro plan. 3.1 Flash Image is deprecated in the release notes and stable on the models page. Embedding 2 has one ID on the models page and another on the deprecations page. Omni is GA on DeepMind and preview on the models page. Name both surfaces and say which one you trusted.

A launch price is a countdown

Argon’s $2 and $10 are introductory, and they end on November 30, 2026. From December 1 it’s $4 and $20, batch included at half. A pilot costed in October, approved in November and rolled out in December gets a 2x bill nobody budgeted. Cost every Argon workload at the December rate.

Rollout fragmentation

Plan, region, language, hardware and Workspace edition can each change access. And the consumer plan tier now decides whether an account reaches Argon, not just how often.

Prompt counts are the wrong unit

Google moved the consumer plans from daily prompt limits to a compute-used model, where limits factor in prompt complexity, the features used and chat length. Any plan comparison built on prompts per day is measuring something Google no longer sells.

Quote the tier with the number

Every published Gemini rate belongs to a tier. Free tier is free of charge on the Flash text models, and not available on the image models, Omni or Argon. Batch is exactly half of paid standard. A batch workload costed at standard overstates spend by 2x.

The cheap model is the one with the end date

Deprecation usually moves you forward and down in price. Not here. The cheapest model, gemini-3.1-flash-lite, at $0.25 per 1M input and $1.50 per 1M output, is scheduled to stop serving on May 7, 2027. Google names the pricier gemini-3.5-flash-lite as its replacement at $0.30 and $2.50. Re-cost any high-volume pipeline before you treat the swap as neutral.

“Legacy” was the warning

Google’s models page called Gemini 3.5 Flash its legacy Flash model. Then, on October 8, it deprecated it and started routing its calls to 3.6 Flash. No shutdown date came first. The reroute was the notice. 3.6 Flash now carries the “previous-generation” label. Plan the move when the label changes, not when the date appears.

Grounding errors

A cited Search result can still be misunderstood, stale or weaker than the claim.

Computer-use risk

An interface can hand your agent instructions you never wrote. Google documents prompt injection detection for Gemini 3.5 Flash-Lite or later, and ships it switched off. So the first risk is a config default. Two more cautions. Argon’s computer use is marked preview. And Agent Anomaly Detection, the runtime oversight layer, was private preview when Google announced it on September 16, 2026.

An agent contract can break under you

Google replaced the default Antigravity agent on September 17, 2026, moving built-in tool parameters to PascalCase and file edits to line ranges. Then it dated the old one to October 5, 2026. That date passed and nobody said what happened. A model swap is visible in a dashboard. A tool-contract change shows up as failing agents, and a shutdown shows up as nothing at all. Read the agent release notes on the same schedule you read the deprecations table.

Background-agent drift

Long-running monitoring needs a clear objective, notification policy and stop condition.

Part of Apple’s private inference runs in Google Cloud

Apple calls its third-generation foundation models “custom-built in collaboration with Google.” And for the most capable, AFM 3 Cloud Pro, it says this: “we worked with Google and NVIDIA to extend Private Cloud Compute to NVIDIA GPUs in Google Cloud.” So some of Apple’s private inference runs on Google’s infrastructure. None of that moves the model and price tables above. But a client may already reach Google through Apple rather than through you. Ask where the model actually runs before you scope a Google agreement.
· Apple Machine Learning Research, June 8, 2026, updated September 9, 2026, on the AFM 3 family, and Apple Security Research, “Expanding Private Cloud Compute,” June 8, 2026.

Personal versus work context

Connected apps and device context can cross boundaries employees or administrators didn’t intend.

Generated media truth

Visual quality doesn’t guarantee accurate text, products, scenes or implied evidence. Image and music output also carry rights questions that a text answer doesn’t.

Product-name churn

Build guidance around durable workflows, with dated model notes, rather than temporary launch labels. NotebookLM became Gemini Notebook on July 16, 2026. Gems become skills from November 2026.

The superlative is a constant

Google called 3.6 Flash, then 3.7 Flash, then 3.8 Flash a version of “our most intelligent workhorse model.” A phrase reused across three generations can’t distinguish any of them. Take the differences from the model card and the price table, and never let a vendor adjective carry a recommendation.

Announced is not available

Spark, the cyber models, enterprise Live access, Argon computer use and generative UI all have to keep their actual rollout state attached. Argon shows how staged a GA can be: paid API only, Antigravity for Ultra, the app and AI Mode for Pro and Ultra, and a Plus label nobody can square with Google’s own post. Spark is the sharpest. Google’s support page, read October 15, 2026, still excludes the European Economic Area, Nigeria, Switzerland and the United Kingdom.

Earliest possible, not final

The deprecations page says its shutdown dates “indicate the earliest possible dates on which a model might be retired,” with advance notice before actual discontinuation. Plan against the published date, but describe it the way Google does. And remember the date can move later too, as gemini-2.5-flash-image did.

A shutdown date can be withdrawn, and access can be cut without one

In early August Google’s deprecations page dated the whole Gemini 2.5 text family to October 16, 2026, then removed it. All three models still read “No shutdown date announced.” But Google has narrowed who can use them: “we are limiting access to the 2.5 models to users who have actively used them in the past.” New projects are pointed at 3.5 Flash-Lite or 3.8 Flash. No shutdown date doesn’t mean a new project can call the model.

A deprecation row can vanish and come back

The preview gemini-omni-flash-preview was dated to stop on September 30, 2026. After that date its row disappeared, even though the page promises that “Already-shutdown models are indicated with gray backgrounds.” Then it reappeared with October 22, 2026. No release note explains either edit. So the deprecations table is neither a history nor a fixed schedule. Keep a dated copy of the row you depend on.

A shutdown date can pass unconfirmed

The preview gemini-robotics-er-1.6-preview was scheduled to shut down on August 31, 2026. As of October 15, 2026, the robotics overview says it “was deprecated on August 31, 2026” and tells anyone “still using” it to upgrade to Robotics ER 2. Deprecated isn’t shut down, and “still using” suggests it may still answer. The models page lists it with no mark. So the status to plan around is a passed date and Google’s wording, not a confirmed shutdown.

A migration target can be on a fuse

The rewritten Imagen page tells users to move to gemini-2.5-flash-image, with gemini-3.1-flash-image in parentheses. The second was deprecated on October 6. The first had its shutdown date moved and its replacement swapped. And the deprecations page now names gemini-nano-banana-2.1 for Imagen users, which the Imagen page never mentions. Check every named replacement against the shutdown table, including the ones a fixed page gives you.

Documentation lags its own product

The Imagen page spent weeks describing an August 17 shutdown in the future tense before Google rewrote it on September 17. So use one rule. A dated article is authoritative for what was announced and when. A live page is authoritative for what is true today. When they disagree, the live page governs the present and the article governs the record. Keep your own dated copy of the live page, because it carries no history.

Whose benchmark is it

Google attributes the 17 percent output-token reduction for 3.6 Flash to the Artificial Analysis Index. The 88 percent token, 66 percent cost and 7 percent accuracy figures for agentic video are Google’s own. So are Argon’s safety figures: all 12 simulated cyberattack scenarios solved, bio red teaming unlocking “less than 1%” of sub-tasks, and a 0.7 percent attack success rate on Gray Swan indirect prompt injection. Keep each figure attached to the party that produced it.

Alias drift

Google’s release notes moved gemini-flash-latest to gemini-3.5-flash on May 19, 2026. That model has routed to 3.6 Flash since October 8, and no note says what the alias does now. Google’s stated policy is a hot swap with two weeks of email notice. So pin an explicit model ID, and treat any code still on the alias as running on an unknown model.

A missing cadence is information

Two consecutive months have closed with no Gemini Drop. July 31, 2026 remains the most recent edition on Google’s blog, and Google’s own Drops hub lists June as its latest. October isn’t due yet, since the July edition landed on the 31st. Don’t present the drop as a reliable monthly channel, and don’t call it dead until the October window closes.

Slipped timelines, plural

Gemini 3.5 Pro was described on May 19, 2026 as rolling out the next month, and on July 21 as testing with partners. DeepMind still shows “3.5 Pro coming soon” in October, with no date and no catalog entry. Gemini 3.1 Pro shipped in preview on February 19, 2026 with general availability promised soon, and it’s still a preview as of October 15, 2026. Meanwhile Gemini 4 Argon went from announcement to GA in 13 days. Two slipped Pro timelines next to a fast flagship tells you where Google’s attention is.

A user-facing feature can be deprecated too

Deprecation isn’t only an API problem. Google replaced Gems with skills on September 30, 2026. Personal accounts lose Gems from November 2026, and Gems move into Settings on November 17. Workspace business, enterprise and nonprofit follow no sooner than March 1, 2027, education no sooner than June 1, 2027. Skills in Workspace apps are delayed until further notice. Inventory every Gem in your organization before November 17.

Gemini claimWhat to confirm before relying on itWhat a dependable version states
Model defaultGoogle’s official model release pageThe surface and the date: “default in the Gemini app as of…”
Model identifierRelease notes for routing and alias changes; the models page for any named replacementAn explicit model ID and its GA date, never an alias
Something Google “hasn’t published”The model card, the pricing page and the API reference, not just one of themWhat the card publishes, with the date the card was read
PriceThe tier, the context tier for Pro, and the end date of any introductory or promotional rateThe tier and the end date in the same sentence as the number
Consumer plan priceThe dated Google post that states the amount; the plans page shows noneThe figure, the date Google stated it and the “price varies by country” caveat
Vendor superlativeWhether the same phrase covered the previous modelThe date and the model the phrase was written about
New agentPlan, region, tester status and the agent tool contract“Rolling out to…” or “available to…”, never a blanket “available”
Workspace featureEdition, admin controls and any delay notice on Workspace UpdatesThe Workspace surface and which editions are eligible
Computer useThe computer-use model list and the prompt injection detection settingThe listed model ID, and whether its support is preview
Creative capabilityModel, product and output availability, plus the per-image rateThe announced model kept separate from the shipping user feature
Model lifespanThe current deprecations page, not a date noted weeks earlierThe earliest possible shutdown date, the named replacement and the date the page was read
A shutdown that already happenedA page saying it happened, not one saying it willThe passed date, plus the Google pages that still disagree
Regional availabilityThe exclusion list, not the launch framingThe excluded regions in the same sentence as the launch
8 / Sources

Where these facts come from

All but one of these Google articles carry a full publication date. The Gemini 3.5 Transcribe announcement is stamped August 2026 with no day, so treat it as undated. Google’s lineup moves across the app, Search and the API at different speeds, so a claim that’s true of one surface isn’t automatically true of another. And much of what a buyer needs lives on pages Google rewrites in place. One of those is listed below because it’s the only record of when a model dies. The rest are named here, each read October 15, 2026, and anyone relying on one should confirm it directly. The pricing page, which carries no stamp at all, for every price in the tables, the Argon introductory and December 1 rates, the Contributor tier and the search grounding wording. The models page, stamped October 13, for model descriptions and listings. The Gemini 4 Argon model page and the “What’s new in Gemini 4 Argon” migration notes. The 3.7 Flash model page and its deprecation banner. The computer-use page, stamped October 13. The robotics overview, stamped October 6. The Imagen page, unchanged since September 17. DeepMind’s Gemini 3.1 Pro page, with its “3.5 Pro coming soon” banner, and its Gemini 3.1 Deep Think page. The Google AI plans page, which shows no prices. The Spark support page. And Google’s Drops hub. Several items exist only in Google’s API release notes, which are rewritten in place: the October 8 deprecation and routing of 3.7 Flash and 3.5 Flash, the October 6 deprecation of 3.1 Flash Image, the September 22 GA of the two 3.8 TTS models, the September 17 Antigravity agent change, the September 3 GA of Lyria 3.5, the August GA dates for Omni 1.1 Flash and 3.5 Transcribe, and the May 19 move of gemini-flash-latest to gemini-3.5-flash.

Gemini 4 Argon: Built to solve your hardest problemsGoogle · October 13, 2026 · general availability of Gemini 4 Argon and where it is availableGemini 4 Argon is now available in the Gemini appGoogle · October 13, 2026 · Argon for Google AI Pro and Ultra subscribers in the Gemini app and AI ModeGemini 4 Argon is now available for developers.Google · October 13, 2026 · Argon through the Gemini API in AI Studio for developers and enterprisesDeveloping Gemini 4 Argon responsibly and safelyGoogle · October 13, 2026 · Google’s own cyber, bio and prompt injection safety figures for ArgonGemini 4 Argon, Model CardGoogle DeepMind · October 13, 2026 · March 2026 knowledge cutoff and the combined 1M-token context windowAn update on the launch of skills in the Gemini app and WorkspaceGoogle Workspace Updates · October 13, 2026 · skills in Workspace apps delayed until further notice; Gemini app rollout completing by mid-NovemberNano Banana 2.1, Model CardGoogle DeepMind · October 6, 2026 · the update to Gemini 3.1 Flash Image and its distribution surfacesThe latest AI news we announced in September 2026Google · October 2, 2026 · Google’s September roundup, including the Gemini app for Windows and Spark on WindowsWorkspace Drops: Pro-level AI image creation and editing, new ways to make your data interactive, and more AI help across your appsGoogle Workspace · October 1, 2026 · the September Workspace Feature Drop, including Gemini data classification in DriveGemini 4 Argon: our next era of frontier intelligenceGoogle · September 30, 2026 · the announcement, the Fairwind-first rollout order, the 1M-token output limit, and a footnote giving $4 and $20 per 1M tokens after November 30, 2026Introducing skills in the Gemini app and Workspace, plus what’s next for GemsGoogle Workspace Updates · September 30, 2026 · Gems move to Settings November 17, 2026; Workspace retirement no sooner than March 1 and June 1, 2027Let skills in Gemini tackle your most repetitive tasksGoogle · September 30, 2026 · skills replace Gems, removal for personal accounts starting in November, and automatic conversion of existing GemsIntroducing Gemini 3.8 Live with Live AvatarGoogle · September 24, 2026 · Live Avatar shipping in Gemini Enterprise, with lip-sync across 97 languages, asynchronous tool execution and SynthID watermarkingGemini API model deprecationsGoogle for Developers · rolling reference, stamped October 12, 2026, checked October 15, 2026 · the only first-party record of shutdown dates and named replacements; rows move and reappear without noticeAgent Anomaly Detection, now in Private Preview on the Gemini Enterprise Agent PlatformGoogle for Developers · September 16, 2026 · trace-based runtime oversight for agentsIntroducing Gemini 3.8 Live and 3.8 Live Extended ThinkingGoogle · September 15, 2026 · general availability of the two audio-to-audio Live API models, including Workspace rolloutBuild zero-trust AI agents that judge intent, not just syntaxGoogle for Developers · September 15, 2026 · runtime agent governanceGemini 3.8 Flash model cardGoogle DeepMind · September 2, 2026 · the March 2026 knowledge cutoff, up to 1M context and 64K output; unconfirmed as of October 15, 2026Gemini 3.8 Flash and Gemini 3.8 Flash CyberGoogle · September 2, 2026 · Gemini 3.8 Flash and 3.8 Flash Cyber general availabilityProactive cyber defense for governments and enterprisesGoogle · September 2, 2026 · the Fairwind Program, the limited-access program that gates the cyber modelIntroducing agentic video understanding in GeminiGoogle · September 1, 2026 · agentic video understanding, named for 3.7 Flash, 3.6 Flash and 3.5 Flash-LiteGemini Omni 1.1 Flash lets you build with more controlGoogle · August 27, 2026 · the dated article behind Omni 1.1; it says production-ready and rolling out, never GAIntelligent transcription with Gemini 3.5 TranscribeGoogle · August 2026 · the article behind the Gemini 3.5 Transcribe modelsCollege students get 12 months of Google AI freeGoogle · August 19, 2026 · the student offer, and the first-party statement of Google AI Pro at $19.99 a monthIntroducing Gemini 3.7 FlashGoogle · August 13, 2026 · the launch of the model deprecated on October 8, with the same workhorse framing later used for 3.8New connected apps and services in the Gemini appGoogle · August 12, 2026 · new connected apps and servicesGoogle’s Gemini app hits 1 billion monthly active usersGoogle · August 11, 2026 · the 1 billion monthly active user figureThe July Gemini DropGoogle · July 31, 2026 · Spark “going global,” and the most recent Gemini Drop on Google’s blogExpanding managed agents in the Gemini APIGoogle · July 28, 2026 · managed agents open to free-tier projects, and the source of the May 2026 Antigravity agent defaultGemini at Galaxy Unpacked 2026Google · July 22, 2026A message from our CEO: Alphabet Q2 2026 earningsGoogle · July 22, 2026 · AI Mode monthly users and tokens per minuteGemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash CyberGoogle · July 21, 2026NotebookLM becomes Gemini NotebookGoogle · July 16, 2026Gemini Omni clip generation and personal avatars in Google VidsGoogle · July 16, 2026Introducing computer use in Gemini 3.5 FlashGoogle · June 24, 2026 · the original announcement, since extended to newer modelsGemini 3.5: frontier intelligence with actionGoogle · May 19, 2026 · the only dated statement of a Gemini 3.5 Pro timelineThe Gemini app becomes more agenticGoogle · May 19, 2026Everything new in our Google AI subscriptions, fresh from I/O 2026Google · May 19, 2026 · the move to a compute-used model, the top Ultra tier cut from $250 to $200 a month, and the new $100 a month Ultra tierGoogle Search I/O 2026 updatesGoogle · May 19, 2026Gemini 3.1 Pro: a smarter model for your most complex tasksGoogle · February 19, 2026 · the launch of the shipping Pro model, released in preview with general availability promised soonGoogle AI Plus is now available everywhere our AI plans are available, including the U.S.Google · January 27, 2026 · the Google AI Plus price at $7.99 a month in the US, with Google’s caveat that price varies by country

AI Mindset

Explore the model cheatsheets