This page collects every model release from our trackers in one table. Changes to a model inside a product are included.
The newest entry is dated 4th Sep 2026, on the AI Mode tracker.
Releases so far by tracker: AI Mode (7), Gemini (5), Grok (4), ChatGPT (2), Microsoft Copilot (1), AI Overviews (1).
A row appears when a model reaches a product we track. Research previews and API-only releases are listed only when they change what a product does.
Model releases and model changes, newest first
Each row is a dated entry on one of our trackers. The link opens the entry with its primary source.
| Date | Event | Tracker | What happened |
|---|---|---|---|
| 2026 | |||
| Google confirms Gemini 3.8 Flash in AI Mode is dropping citations, and promises a fix | AI Mode | Testers found AI Mode on Gemini 3.8 Flash answering broad queries with no links at all. Robby Stein of Google Search said it "isn't working as intended" and a fix is coming. | |
| Lyria 3.5 brings music templates and longer tracks to the Gemini app for every user | Gemini | Google's Lyria 3.5 music model is in the Gemini app for all users worldwide. It adds genre and vocal choices, templates and a short-or-long track option. | |
| GPT-6 Astra starts rolling out to paid ChatGPT tiers, with access off by default for enterprises | ChatGPT | OpenAI released GPT-6 Astra on 3rd Sep to a limited set of organizations. Paid tiers follow "over the coming days"; the free tier is not named. | |
| Gemini 3.8 Flash becomes selectable in AI Mode, three weeks after 3.7 | AI Mode | Gemini 3.8 Flash is now a menu option in AI Mode for AI Pro and Ultra subscribers worldwide. The default model is unchanged, and 3.7 Flash stays paid-only. | |
| Gemini 3.8 Flash reaches the Gemini app for Pro and Ultra subscribers | Gemini | Google's third Flash model in six weeks reaches the Gemini app, AI Mode and Sheets for paid plans. Its predecessor reached the app only through Spark. | |
| Grok 4.6 reaches GitHub Copilot and Amazon Bedrock in five days | Grok | xAI put its flagship into GitHub Copilot on 14th Aug and into general availability on Amazon Bedrock on 19th Aug. Both routes charge 2 dollars per million input tokens and 6 per million output. | |
| Gemini 3.7 Flash becomes a selectable model in AI Mode | AI Mode | Gemini 3.7 Flash is now selectable in AI Mode for AI Pro and Ultra subscribers, globally in English. It is an extra menu option, not a new default. | |
| Gemini 3.7 Flash takes over Gemini Spark | Gemini | Google's new coding and agent model now powers Gemini Spark, for AI Pro and Ultra subscribers in over 160 countries. It is a developer and enterprise model that reaches consumers through exactly one door. | |
| Grok 4.6 becomes the flagship, aimed at long-running agents | Grok | xAI released Grok 4.6, aimed at long-running agents and more ambitious interactive and visual work. It claims parity with GPT-5.6 Sol on a composite benchmark index and went live in Cursor and Grok Build the same day. | |
| GPT-5.6 Luna becomes the free-tier default and free text chats go unlimited | ChatGPT | OpenAI made GPT-5.6 Luna the Free and Go default, replacing GPT-5.5, and announced unlimited text chats for free users. The ads-funded free tier loses its ceiling, which is the point of the ads. | |
| Build Mode ships, and Grok 4.5 arrives in GitHub Copilot | Grok | Build Mode builds a working site, app or dashboard in the chat, publishing to a grok.me link or custom domain. The same day, Grok 4.5 became selectable in GitHub Copilot's model picker. | |
| Grok 4.5 reaches grok.com, X, iOS and Android | Grok | xAI rolled Grok 4.5 out across every consumer surface as its most capable general model. It is framed for everyday questions and for knowledge work like spreadsheets, decks and drafting. | |
| Gemini 3.5 Flash-Lite starts rolling out in Google Search | AI Mode | Google announced Gemini 3.5 Flash-Lite, its fastest and cheapest 3.5-class model, "also rolling out in Google Search". Google did not say what it powers there, and AI Mode still runs on Gemini 3.5 Flash. | |
| 2025 | |||
| Gemini 3 Flash becomes the default AI Mode model worldwide | AI Mode | Gemini 3 Flash became the default AI Mode model globally. Gemini 3 Pro opened to everyone in the US through the Thinking with 3 Pro picker. | |
| Gemini 3 Flash becomes the app default, with a Fast, Thinking and Pro picker | Gemini | Gemini 3 Flash replaced 2.5 Flash as the Gemini app's default model, with a picker labelled Fast, Thinking and Pro. Those labels have since given way to model names plus a separate thinking-level control. | |
| Gemini 3 Pro in AI Mode expands to nearly 120 countries | AI Mode | Gemini 3 Pro in AI Mode reached nearly 120 countries and territories in English for AI Pro and Ultra subscribers. That came two weeks after its US debut. | |
| Gemini 3 Pro lands app-wide and Gemini Agent arrives | Gemini | Gemini 3 Pro reached the whole Gemini app alongside Gemini Agent, its first agentic feature and the ancestor of Spark. It is the point where the assistant started doing things rather than answering questions. | |
| GPT-5 arrives in Copilot with Smart Mode routing | Microsoft Copilot | Consumer Copilot got GPT-5 on the model's launch day, across all markets and platforms. A Smart Mode routes each query to the right model variant. | |
| Gemini 2.5 Pro and Deep Search arrive for AI Pro and Ultra subscribers | AI Mode | US AI Pro and Ultra subscribers got a Gemini 2.5 Pro picker and Deep Search in AI Mode via Labs. Agentic calling of local businesses began rolling out to US Search users. | |
| AI Overviews upgraded to Gemini 2.0 and opened to teens and signed-out users | AI Overviews | Google rolled Gemini 2.0 into US AI Overviews for harder questions, and opened the feature to teens and signed-out users. It introduced AI Mode as a Search Labs experiment the same day. | |
What counts as a release here
A model matters to this section when it changes what a search or assistant product does.
We track products, not laboratories. A model earns a row when it becomes the default, becomes selectable, or arrives in a product we cover.
A change in how a model behaves inside a product, such as a citation fault, is also a row. The type of each entry on its tracker is "Model".
The trackers are ChatGPT, Gemini, AI Mode, AI Overviews, Claude, Perplexity, Microsoft Copilot and Grok.
Some labs have no product we track: DeepSeek, Llama and Mistral among them. Their models appear only when they land inside a tracked product.
Every row links to the entry, and every entry links to the announcement it rests on.
Reading the table
The date is the day the model reached the product, not the day the lab first described it.
The date column is the entry date: the day the change became visible in the product. A model announced in a blog post and shipped a week later is dated to the shipping.
The tracker column says which product the row is about. The same model can appear twice when it reaches two products on different days.
The summary is the entry's own summary, unchanged.
Frequently asked questions
What is the latest AI model release?
The most recent model entry is dated 4th Sep 2026, on the AI Mode tracker. The table above updates whenever a tracker records a new one.
How many AI models were released in 2026?
Our trackers record 13 model releases in 2026 so far. That counts releases into products we cover, not every model any lab published.
Why is a model I know about missing?
Either it never reached a product we track, or it arrived before that tracker existed. Our AI assistant trackers start in 2025.
Email a tip through the box below if a release inside a tracked product is missing.
Compiled by Claude from primary sources, checked by Akash Mansukhani before publication. See how we source and correct this page.Keywords Everywhere, "AI model releases", last updated 6th Sep 2026, https://keywordseverywhere.com/news/ai-model-releases/