Start hereOpenAI's model index says that if you're not sure where to start, use GPT-6 Astra. Anthropic recommends Claude Opus 5.5 for most workloads. So who makes the models people compare, and which are open?
The table
The companies that build large language models are often called AI labs. The table lists the main ones, alphabetically, with each one’s current top model as read from the labs’ own documentation on 21 October 2026. Model line-ups change every few weeks, so treat this as a dated snapshot.
Context window is the figure each lab publishes for how many tokens a model can handle at once. Labs do not all count it the same way: for some it includes the model’s reply as well as your input.
| Lab | Current top model | Released | Weights | Context window |
|---|---|---|---|---|
| Alibaba (Qwen) | Qwen3.8-Max; weights published as Qwen3.8-2.4T-A95B | API: 3 August 2026 | Downloadable, custom licence | 262,144 natively, extensible to 1,010,000 (weights) |
| Anthropic | Claude Fable 5.1; Claude Opus 5.5 | 1 September 2026; 22 September 2026 | Closed | 1M tokens |
| DeepSeek | DeepSeek-V4-Pro; DeepSeek-V4.1-Flash-1017 | 13 August 2026 (V4-Pro general release); 17 October 2026 (V4.1-Flash update) | Downloadable, MIT | 1M tokens |
| Google DeepMind | Gemini 4 Argon; Gemini 3.8 Flash | 13 October 2026; 2 September 2026 (both general availability) | Closed | 1,048,576 |
| Meta | Muse Wonder; Muse Wonder Full | 12 October 2026 | Closed | 1M tokens |
| Mistral AI | Mistral Large 4 (public preview); Mistral Medium 3.5 | 6 October 2026; April 2026 | Large 4: weights promised for the end of October; Medium 3.5: downloadable, custom licence | 1M; 256k |
| Moonshot AI | Kimi K3 | July 2026 | Downloadable, custom licence | 1,048,576 |
| OpenAI | GPT-6 Astra | 3 September 2026 | Closed | 1,050,000 |
| xAI (SpaceXSI) | Grok 4.7 | API: 21 September 2026 | Closed | 500k |
| Z.ai (Zhipu) | GLM-5.3 | 18 August 2026 | Downloadable, custom licence | 1M tokens |
What the labs say about their own models
These are the labs’ own descriptions, not independent assessments.
- Alibaba: its model card says Qwen3.8 is the first time a Qwen-Max-class model has had an open release, with 2.4 trillion parameters in total and 95 billion activated.
- Anthropic: its documentation recommends Claude Opus 5.5 for most workloads, and Claude Fable 5.1 for demanding reasoning and long-horizon agentic work. Anthropic also launched Claude Sonnet 5.5 on 28 September and Claude Haiku 5.5 on 7 October 2026.
- DeepSeek: its model card calls the 0813 release the official V4-Pro, replacing the preview with much stronger agent abilities.
- Google: on 13 October 2026 it made Gemini 4 Argon generally available, calling it its most capable model, built for complex coding, agentic, math and science tasks. It calls Gemini 3.8 Flash its most intelligent Flash model, built for long-running software engineering, autonomous agents and complex enterprise workflows. Gemini 3.1 Pro is still listed as a preview, with no shutdown date announced.
- Meta: on 12 October 2026 it introduced Muse Wonder and Muse Wonder Full, which it calls “our most capable models to date”. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is for Muse Code subscribers. Its earlier Muse Spark line came first. At launch in April 2026 Meta said it hoped to open-source future versions. Announcing version 1.3, it listed a Muse Spark open weights release on its roadmap. On 15 October 2026 it released the weights for Muse Spark Open, which it calls the most capable US open-weight model to date, under its Muse Model License “with limited commercial terms”.
- Mistral AI: on 6 October 2026 it launched a public preview of Mistral Large 4, which it calls its largest and most capable model to date. It says: “We will release the weights by the end of the month.” Its model page lists the licence only as “Open”. Mistral Medium 3.5, from April 2026, was “Released as open weights under a Modified MIT license.”
- Moonshot AI: its model card calls Kimi K3 an open-weight, native multimodal agentic model and its most capable model to date.
- OpenAI: its model index says “If you’re not sure where to start, use GPT-6 Astra”. It suggests GPT-6.1 Sol to balance intelligence and cost, and GPT-6 Luna for cost-sensitive, high-volume work. Its system card for GPT-6 Astra is dated 3 September 2026.
- xAI: SpaceX announced on 2 February 2026 that it had acquired xAI. xAI’s own pages now use the name SpaceXSI, with no announcement on its news page; the copyright line reads SpaceXAI LLC. Its models page calls Grok 4.7 its flagship model for code and everything else.
- Z.ai: its guide says GLM-5.3 takes text-only input, with a 1M-token context window and a maximum output of 128K tokens.
Closed or downloadable?
A closed model’s weights are not published: you use it through an app or an API. Some closed models are also sold through other companies’ cloud platforms, such as Amazon Bedrock or Microsoft Foundry.
A model with downloadable weights can be run on hardware you control, but downloadable is not the same as open-source. The licence decides what you may do with it. Several of the downloadable models above use custom licences with extra conditions for large companies. See Open-weight is not open-source for what those licences say.
Some labs publish both kinds: Meta’s Muse Glimmer model (30B parameters), for example, is released with open weights under Apache 2.0.
Models get retired
Labs publish schedules for switching old models off, and those dates matter if you build on a model:
- Anthropic’s deprecation page says Claude Opus 4 and Claude Sonnet 4 were retired on 15 June 2026.
- OpenAI says it notified developers of older GPT-5 and o3 snapshots of their deprecation and removal from the API on 11 December 2026.
- Google’s deprecation page notes: “The shutdown dates listed in the table indicate the earliest possible dates”.
- On 8 October 2026 Google deprecated Gemini 3.7 Flash, and requests to it are now routed automatically to Gemini 3.8 Flash.
- Plans can also change. DeepSeek’s change log says it decided to keep providing API services for DeepSeek V4 Pro after 14 September 2026.
| 15 June 2026 | Anthropic retires Claude Opus 4 and Claude Sonnet 4 |
| 14 September 2026 | DeepSeek decides to keep V4 Pro’s API running after this date |
| 8 October 2026 | Google deprecates Gemini 3.7 Flash and routes it to Gemini 3.8 Flash |
| 11 December 2026 | OpenAI’s removal date for older GPT-5 and o3 snapshots |
How to check what is current
Each lab keeps an official model page. The ones used for this table were:
- OpenAI: developers.openai.com/api/docs/models
- Anthropic: platform.claude.com/docs/en/about-claude/models/overview
- Google: ai.google.dev/gemini-api/docs/models
- xAI: docs.x.ai/developers/models
- Meta: developer.meta.com/ai/models
- DeepSeek: api-docs.deepseek.com
- Alibaba Qwen: huggingface.co/Qwen and docs.qwencloud.com
- Mistral AI: docs.mistral.ai/models
- Moonshot AI: huggingface.co/moonshotai
- Z.ai: docs.z.ai
Our reading: the table tells you who makes what, not which model is best. Labs’ descriptions of their own models are not independent assessments, and published scores depend heavily on how models are tested. See AI benchmarks explained. The practical test is how a model does on your own tasks.
Quick answers
Which AI model is best right now?
This table does not rank models. Which is best depends on the task, the cost and how it is tested, and labs' descriptions of their own models are not independent assessments.
Is Mistral Large 4 open-weight?
Mistral calls it open-weight, but on 6 October 2026 it launched only a public preview. It says it will release the weights by the end of the month, and its model page lists the licence only as "Open".
What happened to xAI?
SpaceX announced on 2 February 2026 that it had acquired xAI. xAI's own pages now use the name SpaceXSI, and its copyright line reads SpaceXAI LLC; we found no announcement of the change on its news page.
Which leading AI models have downloadable weights?
In this table: DeepSeek-V4-Pro (MIT licence), Qwen3.8, Mistral Medium 3.5, Kimi K3 and GLM-5.3. Four of the five use custom licences, several with extra conditions for large companies. The top models from Anthropic, Google, Meta, OpenAI and xAI are closed.
Do old AI models stop working?
Yes, on published schedules. Anthropic retired Claude Opus 4 and Claude Sonnet 4 on 15 June 2026, and OpenAI plans to remove older GPT-5 and o3 snapshots on 11 December 2026.
Use this
Check a model's own page, and the date you read it, before you build on it.
Is this still the lab's current model, and when will it be retired?
Quick quiz
Nothing is scored or saved. Recalling it once now is what makes it last.
1. Which Meta models were introduced on 12 October 2026?
Answer: Muse Wonder and Muse Wonder Full.Read it again ↑
2. What does downloadable weights mean here?
Answer: You can run it on hardware you control, under its licence.Read it again ↑
3. What does Google's deprecation page say its shutdown dates are?
Answer: The earliest possible dates.Read it again ↑
How we checked this · Last checked October 2026
- Every row was read from the lab's own pages on 21 October 2026: model indexes, changelogs, release notes, deprecation pages and model cards.
- Release dates are the labs' own. Where a lab gives only a month, the table gives the month.
- Labs count context windows differently; each figure is the lab's own.
- Mistral Large 4's weights were not yet published when we read its pages, so its licence is not yet known.
- The page does not list every lab. It covers the main ones and shows who makes what, not which model is best.
Sources
- OpenAI: API models index; API changelog; GPT-6 Astra model page; GPT-6 Astra System Card; deprecations.Primary
- Anthropic: models overview; release notes; model deprecations.Primary
- Google: Gemini API models, changelog and deprecations; Gemini 4 Argon, Gemini 3.8 Flash and Gemini 3.1 Pro model pages.Primary
- xAI: API models and release notes; “xAI joins SpaceX” (2 February 2026), x.ai.Primary
- Meta: “Introducing Muse Wonder and Muse Wonder Full” (12 October 2026), about.fb.com; “Introducing Muse Spark” (8 April 2026), about.fb.com; “Introducing Muse Spark 1.3” (2 September 2026), research.meta.ai; “Introducing Muse Spark Open” (15 October 2026), research.meta.ai; Meta Model API, Muse Wonder and Muse Glimmer pages, developer.meta.com.Primary
- DeepSeek: API news, change log and pricing pages; DeepSeek-V4-Pro-0813 model card, Hugging Face.Primary
- Alibaba Qwen: Qwen Cloud model changelog; Qwen3.8-2.4T-A95B model card, Hugging Face.Primary
- Mistral AI: “Introducing Mistral Large 4” (6 October 2026); Mistral Large 4 model page; Mistral Medium 3.5 documentation; Mistral-Medium-3.5-128B model card, Hugging Face.Primary
- Moonshot AI: Kimi-K3 model card, Hugging Face.Primary
- Z.ai: release notes; GLM-5.3 guide.Primary
Related: What is a frontier AI model? · How large language models work