Home › Companies & Models
Hub · Companies & Models
Companies & Models: who builds the AI
The labs behind today’s chatbots, the models they list on their own pages, how you can reach them, and which ones can be downloaded. Sourced to each lab’s official page.
Last reviewed Oct 3, 2026
Last reviewed Oct 3, 2026. Names, versions and prices change often; the date on each source is the day we read it.
How to read this page
A lab is a company that builds AI models. A model is the trained system (for example GPT-6 Astra); a product is the app you use (for example ChatGPT). Labs rename and retire models often, so this page lists only what each lab’s own page showed on Oct 3, 2026 and links to it. It is not a ranking: we have not tested these models or compared their quality.
Prices are per million tokens (MTok) on each lab’s developer page; what you pay for a chat app is a separate subscription. Terms: tokens, open vs closed, context window.
Labs and models
| Lab | Main products | Current models on its own page | Access and openness (as the page says) |
|---|---|---|---|
| OpenAI | ChatGPT, Codex | GPT-6 Astra (flagship), GPT-6.1 Sol, GPT-6 Luna. Context window 1.05M tokens. List prices per MTok (input / output): Astra $10 / $50, Sol $2 / $10, Luna $0.10 / $0.50.1 | Hosted models; also two open-weight models, gpt-oss-120b and gpt-oss-20b (Apache 2.0).2 |
| Anthropic | Claude, Claude Code | Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 4.5. Context window 1M tokens (Haiku 4.5: 200K). Per MTok: Fable $10 / $50, Opus $4 / $20, Sonnet $2 / $10, Haiku $1 / $5.3 | Available through the Claude API and apps (per its model docs). |
| Gemini, Gemini Live | Gemini 3.8 Flash (described as its most intelligent Flash model), Gemini 3.1 Pro (preview), Gemini 3.8 Live for voice, and Nano Banana image models, per the Gemini API model list.4 | Available through the Gemini API and apps (per its model list). | |
| xAI | Grok | Grok 4.7 (called its flagship model, 500K-token context; $2 / $6 per MTok), Grok Imagine Image 2.0 and Video 1.5, and a Voice API. xAI’s documentation and home page are titled “SpaceXAI”.5,6 | Available through the xAI API and Grok apps (per its docs). |
| Meta | Meta AI | Muse Spark 1.3 (also 1.2 and 1.1), Muse Image, Muse Voice Transcribe, and SAM 3.1, listed at llama.com.7 Its Hugging Face page also lists the Llama 4 collection.8 | Muse Glimmer is listed as available on Hugging Face; Llama models require accepting a license and acceptable use policy.7,8 |
| Mistral AI | Mistral (Vibe, formerly Le Chat) | Mistral Medium 3.5, Mistral Small 4, Mistral Large 3, Ministral 3 (14B, 8B, 3B), plus OCR, audio and code models.9 | Docs mark Small 4, Large 3 and Ministral 3 as Apache 2.0 open-weight.9 |
| DeepSeek | DeepSeek app | DeepSeek-V4.1-Flash, announced on its home page; the API lists model names deepseek-flash and deepseek-v4-pro.10,11 | Publishes models on Hugging Face (organization deepseek-ai); we did not read license terms for the latest models. |
| Alibaba Cloud (Qwen) | Qwen Chat | The Qwen team’s GitHub lists Qwen3.8, Qwen3-VL and Qwen3-Omni; Hugging Face describes Qwen as the model family built by Alibaba Cloud.12,13 | Models are published on Hugging Face and GitHub; we did not read license terms. |
| Amazon | Amazon Nova (Nova models, Nova Forge, Nova Act) | Nova models, plus Nova Forge (a service to build your own frontier models) and Nova Act (browser agents). AWS says Nova is built on technology first developed for Amazon products such as Alexa+.14 | AWS describes Nova as a proprietary model family it lets customers customize. |
Open-weight releases
Open-weight means the trained model files can be downloaded, as defined in the Open vs closed guide. From the official pages above: OpenAI’s gpt-oss models and several Mistral models are labeled Apache 2.0; Meta lists Muse Glimmer on Hugging Face and its Llama models require accepting license terms; DeepSeek and Qwen publish on Hugging Face. Check each model’s own license before you use it commercially. "Open-weight" is not the same as the Open Source Initiative’s "Open Source AI".2,9,7,8,13
More companies and tools
This page covers the labs that train their own large models. Tools built on them, and many smaller makers, are in the AI Systems directory (A–Z, official links only), including Sesame, a voice AI companion app, and Character.AI. We did not duplicate those entries here. For a plain comparison of the big chatbots (free tiers, prices, privacy), see Which AI?. Where each lab’s data centers are: Data Centers and AI by Country.
Keep reading
What we left out. We did not list benchmark scores or rankings (vendor-reported scores are not comparable, and we have not tested the models), company valuations or revenue (no primary page for all labs), or employee counts. Microsoft, Cohere, Moonshot AI (Kimi), Z.ai and other labs are not on this page yet; some appear in the directory. We did not read OpenAI’s pricing page for ChatGPT plans, Google’s per-model prices or Meta’s license text. See How we verify; report an error through the corrections page.
Sources for this page
- Models | OpenAI API. OpenAI.
- Introducing gpt-oss. OpenAI.
- Models overview. Anthropic (Claude Platform Docs).
- Models | Gemini API. Google AI for Developers.
- Grok Models & Pricing. xAI documentation (pages titled “SpaceXAI”).
- SpaceXAI home page (x.ai). xAI.
- Llama.com: available models. Meta.
- Meta Llama (official Hugging Face organization). Meta.
- Models Overview. Mistral AI (docs).
- DeepSeek home page. DeepSeek.
- Your First API Call. DeepSeek API Docs.
- Qwen organization on GitHub. Alibaba Qwen team.
- Qwen organization on Hugging Face. Alibaba Cloud (Qwen).
- Amazon Nova. Amazon Web Services.