ai model releases calendar beta

A calendar of every notable AI model release — language models, image, video, speech and music generators — from 33 studios, with release dates, context windows, prices and benchmark results. It currently holds 293 models released over the last twelve months and updates on its own as new models ship.

What sets it apart: for each major lab it estimates when the next model is likely to arrive, combining the lab's release history with current rumors, and it collects benchmark results from several independent leaderboards, not one, plus the scores studios report in their model cards.

how we date releases

Every date comes from the lab itself: its API model catalog, official X account, blog or API changelog. When a model first appears in a lab's API, the tracker records the moment it went live; an earlier official announcement replaces that date. Models available only by application or approval are marked as limited access, and models a lab has shut down are removed.

benchmarks

Scores come from Epoch AI (the Epoch Capabilities Index, GPQA Diamond, SWE-bench Verified, FrontierMath, SimpleQA Verified) and from open leaderboards that test new models within days of release: Supabase Evals, Next.js Evals and DeepSWE. Text models also carry the Artificial Analysis Intelligence, Coding and Agentic indexes, marked "AA". Each model shows its place in every test, with the leader's result marked on the bar. Tests are listed from the model's best place to its weakest. A test that most new models already score 95% or more on is not shown: a place there no longer tells models apart.

Image and video models are ranked by LMArena, where people vote between two anonymous outputs of the same prompt: text-to-image, image editing, text-to-video and image-to-video arenas (CC BY 4.0), and image models also by the Design Arena image arena. The calendar shows each model's arena rating and its place among all models in that arena.

Sources and licences: Epoch AI (CC BY 4.0); LMArena (CC BY 4.0); Supabase Evals (Apache-2.0); Next.js Evals (MIT); DeepSWE. Source: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings), CC BY 4.0. Source: Design Arena (www.designarena.ai) via OpenRouter (openrouter.ai/rankings), CC BY 4.0.

Many open models are not tested by these leaderboards yet. For them the calendar also shows the results the studio published in the model card on Hugging Face, signed "reported by" the studio and listed after the independent scores. Such a result is the studio's own claim; when an independent score for the same test exists, the calendar shows that one instead.

frequently asked questions

What is the latest AI model release?

The most recent release in the calendar is FLUX 3 Image by Black Forest Labs, released on October 1, 2026. Before it: Grok 4.7 (xAI, Oct 1), Grok Imagine Video 1.5 Lite (xAI, Oct 1), MAI-Voice-2.1 (Microsoft, Oct 1), MAI-Voice-2.1-Flash (Microsoft, Oct 1).

When is the next OpenAI model expected?

Our estimate for the next OpenAI release is October 4–13, 2026, based on the studio's release rhythm and current rumors. The chance of a release within 30 days is about 97%. It is an estimate, not an announcement, and it updates every day.

When is the next Anthropic model expected?

Our estimate for the next Anthropic release is October 8–31, 2026, based on the studio's release rhythm and current rumors. The chance of a release within 30 days is about 76%. It is an estimate, not an announcement, and it updates every day.

When is the next Google model expected?

Our estimate for the next Google release is October 3–11, 2026, based on the studio's release rhythm and current rumors. The chance of a release within 30 days is about 99%. It is an estimate, not an announcement, and it updates every day.

Which AI model leads the benchmarks?

On GPQA Diamond the leader among the models in the calendar is GPT-6 Astra with 95.8%. On SWE-bench Verified the leader among the models in the calendar is Claude Opus 4.7 with 83.5%. On FrontierMath T1–3 the leader among the models in the calendar is GPT-6 Astra with 93.7%. Every model window in the calendar shows its place in each test.

What are the newest AI image models?

The latest image models in the calendar: FLUX 3 Image by Black Forest Labs (Oct 1), Seedream 5.0 Flash by ByteDance Seed (Oct 1), GPT Image 2.5 Flare by OpenAI (Sep 8).

What are the newest AI video models?

The latest video models in the calendar: Grok Imagine Video 1.5 Lite by xAI (Oct 1), Praxis-1 by Runway (Sep 30), Kling 4.0 Flash by Kling (Sep 28).

What are the newest AI music models?

The latest music models in the calendar: Suno v6 by Suno (Sep 9), Suno v6-mini by Suno (Sep 9), Suno v6-wild by Suno (Sep 9).

How is the next release estimated?

From two things: how often the studio has shipped models before (the typical gap between its releases) and dated rumors and teasers about its next model. Official teasers weigh more than rumors, and fresh signals more than old ones. The result is a date range with a chance of a release within 30 days.

Where do the benchmark scores come from?

Independent scores come from Epoch AI, the Artificial Analysis indexes and open leaderboards (Supabase Evals, Next.js Evals, DeepSWE); image and video models are ranked by the LMArena arenas and Design Arena. For models they have not tested, the calendar shows the results the studio itself published in the model card, marked "reported by" the studio. An independent score always replaces a reported one for the same test.

Where do the release dates come from?

From official sources only: the model catalogs of the labs' APIs, their official X accounts and blogs, and their API changelogs. A model is dated when it became publicly available; models open only by application are marked as limited access.

Is the AI releases calendar free?

Yes. The calendar is free, has no sign-up and updates on its own every few minutes.