From the model makers' own channels
Latest AI model releases
New models announced by OpenAI, Anthropic, Google and Mistral, and new open-weight models published on the official Hugging Face accounts of Meta, Mistral, Google, Qwen, DeepSeek and others. Each item links to the original source.
Updated · refreshed twice a day
Official model announcements
From the last four months, newest first.
- Google DeepMind18 Nov 2026
Introducing Gemini 4 Neon (opens Google DeepMind in a new tab)
- Google DeepMind18 Nov 2026
Genie 4 is now available (opens Google DeepMind in a new tab)
- OpenAI17 Nov 2026
Introducing GPT-6.2 Astra and Sol (opens OpenAI in a new tab)
Introducing GPT‑6.2 Astra and Sol, with better professional outputs, coding, computer use, and efficiency at lower cost across ChatGPT Work, Codex, and the API.
- Google DeepMind12 Nov 2026
Introducing Gemini 4 Hydrogen (opens Google DeepMind in a new tab)
- Google DeepMind10 Nov 2026
Gemini 4 Carbon is now available (opens Google DeepMind in a new tab)
- Anthropic9 Nov 2026
Introducing Claude Fable 5.2 and Claude Mythos 5.2 (opens Anthropic in a new tab)
- Google DeepMind28 Oct 2026
Gemini 4 Carbon: Our next frontier model (opens Google DeepMind in a new tab)
- Google DeepMind27 Oct 2026
Genie 4: A new generation of world simulation (opens Google DeepMind in a new tab)
Genie 4: our new world model unlocking unprecedented realism in real-time simulation.
- OpenAI26 Oct 2026
Introducing GPT-6.1 Terra (opens OpenAI in a new tab)
Delivering fast and capable computer use at scale
- Mistral AI20 Oct 2026
Voxtral 3 (opens Mistral AI in a new tab)
Voxtral 3: Mistral's voice agent SDK for real-time speech in 50+ languages, accurate transcription, emotional TTS, deep tool use.
- Mistral AI15 Oct 2026
Introducing Timestral (opens Mistral AI in a new tab)
State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready.
- Google DeepMind13 Oct 2026
Gemini 4 Argon for Robotics (opens Google DeepMind in a new tab)
New open-weight models
Published in the last 45 days on the model makers' official Hugging Face accounts. Open-weight models can be downloaded and run on your own hardware, under each model's licence.
Microsoft
Aion-1.0 (opens Hugging Face in a new tab)
Text / chat
18 Nov 2026 · ♥ 191
NVIDIA
Qwen3.5-Qwen3.8-Draft (opens Hugging Face in a new tab)
Text / chat
11 Nov 2026 · ♥ 179
Microsoft
Mage-VL-Cowork (opens Hugging Face in a new tab)
Vision + text
30 Oct 2026 · ♥ 231
NVIDIA
MotionBricks (opens Hugging Face in a new tab)
Model
23 Oct 2026 · ♥ 139
DeepSeek
DeepSeek-V4.1-Flash-1017 (opens Hugging Face in a new tab)
Vision + text
17 Oct 2026 · ♥ 5,813
IBM
granite-switch-4.1 (opens Hugging Face in a new tab)
Text / chat
14 Oct 2026 · ♥ 56
Mistral AI
Timestral (opens Hugging Face in a new tab)
Model
14 Oct 2026 · 3 versions · ♥ 649
Alibaba Qwen
Qwen-Image-2.1 (opens Hugging Face in a new tab)
Image generation
9 Oct 2026 · ♥ 1,152
Mistral AI
Voxtral-Mini-Realtime-Arabic (opens Hugging Face in a new tab)
Speech to text
8 Oct 2026 · ♥ 56
How to use this list
- A new model isn't automatically better for your job. Test it on your own examples before switching (see Lessons 25 and 44 of the AI Skills course).
- Check the price, speed, context length, data terms and, for open-weight models, the licence.
- Only the model makers' own announcements and official accounts are used, so third-party copies and re-uploads are left out.
Sources
Learn the toolkit
Every framework for building AI applications, how to use it, and which model to use for which job, in plain English.