The model that reached a third of Vercel's paying teams fastest cannot write a sentence
Jev picks between a small set of predetermined outputs rather than generating text. Vercel's figures cover its own gateway, not the labs' direct APIs.
AI Models
Track every significant AI model release with context windows, modalities, availability and benchmark results, updated as new versions ship.
Models
| Model | Developer | Released | Context window | Status |
|---|---|---|---|---|
| Glimpse | Elorian | October 16, 2026 | — | Active |
| FLUX 3 Image Klein | Black Forest Labs | October 15, 2026 | — | Active |
| MAI-Base-2 | Microsoft | October 15, 2026 | — | Announced |
| Muse Spark Open | Meta AI | October 15, 2026 | 1,000,000 | Active |
| Nemotron 3.5 Super VL | NVIDIA | October 15, 2026 | — | Active |
| Grok Voice 3.0 | SpaceXAI | October 14, 2026 | — | Active |
| Muse Wonder | Meta AI | October 12, 2026 | — | Active |
| Muse Wonder Full | Meta AI | October 12, 2026 | — | Active |
| Clef-omni | Cloudflare | October 9, 2026 | 64,000 | Active |
| Microsoft-Decision-1 | Microsoft | October 9, 2026 | — | Active |
| Odyssey-3 | Odyssey | October 8, 2026 | — | Active |
| Claude Haiku 5.5 | Anthropic | October 7, 2026 | — | Active |
| EmbeddingGemma 2 | October 6, 2026 | — | Active | |
| Mistral Large 4 | Mistral AI | October 6, 2026 | — | Announced |
| Nano Banana 2.1 | October 6, 2026 | 1,000,000 | Active | |
| Amazon Nova 2.5 Sonic | Amazon | October 5, 2026 | 256,000 | Active |
| Beam | Reflection AI | October 5, 2026 | 1,000,000 | Announced |
| Rho-1 | Reka AI | October 5, 2026 | — | Announced |
| Kolibri | Aleph Alpha | October 3, 2026 | 1,000,000 | Active |
| MAI-Transcribe-2-Streaming | Microsoft | October 1, 2026 | — | Active |
| MAI-Voice-2.1 | Microsoft | October 1, 2026 | — | Active |
| pplx-decider-v1-27b | Perplexity | October 1, 2026 | — | Active |
| Gemini 4 Argon | Google DeepMind | September 30, 2026 | — | Active |
| Phonon-2 | Fermion Research | September 30, 2026 | — | Active |
| GPT-6.1 Sol | OpenAI | September 29, 2026 | — | Active |
| Claude Sonnet 5.5 | Anthropic | September 28, 2026 | — | Active |
| Naive-N0.5-Flash | Naive AI | September 27, 2026 | 1,000,000 | Active |
| FLUX 3 Action | Black Forest Labs | September 24, 2026 | — | Active |
| CLM-8B | Stanford Hazy Research | September 23, 2026 | — | Active |
| Gemini 3.8 Flash TTS | September 23, 2026 | — | Active | |
| Gemini 3.8 Flash-Lite TTS | September 23, 2026 | — | Active | |
| Nemotron 3 Diarization | NVIDIA | September 23, 2026 | — | Active |
| Qwen-Audio-3.1 | Alibaba | September 23, 2026 | — | Active |
| Claude Opus 5.5 | Anthropic | September 22, 2026 | — | Active |
| GPT-6 Luna | OpenAI | September 22, 2026 | — | Active |
| GPT-6 Sol | OpenAI | September 22, 2026 | — | Active |
| Grok 4.7 | SpaceXAI | September 21, 2026 | — | Active |
| MiMo-V2.6-Pro | Xiaomi | September 21, 2026 | 1,000,000 | Active |
| Qwen-Image-2.1 | Alibaba | September 20, 2026 | — | Active |
| Step 5 Preview | StepFun | September 20, 2026 | 1,000,000 | Announced |
| Grok Voice Transcribe 2.0 | SpaceXAI | September 18, 2026 | — | Active |
| Qwen3.8-Omni-Flash | Alibaba | September 18, 2026 | 1,000,000 | Active |
| Bonsai 2 27B | PrismML | September 17, 2026 | 262,144 | Active |
| Gemini 3.8 Live | September 15, 2026 | — | Active | |
| Jev | TypeSafe AI | September 15, 2026 | — | Active |
| Atria Dawn Preview | Shanghai AI Laboratory | September 11, 2026 | 262,144 | Active |
| DeepSeek-V4.1-Flash | DeepSeek | September 10, 2026 | 1,000,000 | Active |
| GPT-Live-1 | OpenAI | September 10, 2026 | — | Active |
| Qwen-Drive-1.0 | Alibaba | September 7, 2026 | — | Active |
| GPT-6 Astra | OpenAI | September 3, 2026 | 1,050,000 | Active |
| humain-m3 | HUMAIN | September 3, 2026 | — | Announced |
| K2 Horizon | Institute of Foundation Models | September 3, 2026 | — | Active |
| MAI-Transcribe-2 | Microsoft | September 3, 2026 | — | Active |
| WeatherNext 3 | Google DeepMind | September 3, 2026 | — | Active |
| Muse Voice Transcribe | Meta AI | September 1, 2026 | — | Active |
| GLM-5.3-Flash | Zhipu AI | August 26, 2026 | 1,000,000 | Active |
| Lyria 3.5 | Google DeepMind | July 29, 2026 | — | Active |
| Kimi K3 | Moonshot AI | July 1, 2026 | — | Active |
| Claude Mythos | Anthropic | April 1, 2026 | — | Active |
| GPT-5 | OpenAI | August 7, 2025 | 400,000 | Active |
| Claude Opus 4.1 | Anthropic | August 5, 2025 | 200,000 | Active |
| Kimi K2 | Moonshot AI | July 11, 2025 | 128,000 | Active |
| Grok 4 | SpaceXAI | July 9, 2025 | 256,000 | Active |
| Claude Sonnet 4 | Anthropic | May 22, 2025 | 200,000 | Active |
| Qwen3-235B | Alibaba | April 29, 2025 | 128,000 | Active |
| o3 | OpenAI | April 16, 2025 | 200,000 | Legacy |
| Llama 4 Maverick | Meta AI | April 5, 2025 | 1,000,000 | Active |
| Gemini 2.5 Pro | Google DeepMind | March 25, 2025 | 1,048,576 | Active |
| DeepSeek-R1 | DeepSeek | January 20, 2025 | 64,000 | Active |
| Mistral Large 2 | Mistral AI | July 24, 2024 | 128,000 | Active |
| GPT-4o | OpenAI | May 13, 2024 | 128,000 | Legacy |
| Claude Opus 5 | Anthropic | — | — | Active |
| GLM-5.3 | Zhipu AI | — | 1,000,000 | Active |
Latest
All stories
Glimpse revisits the image while it reasons rather than converting it to tokens once. Elorian's own table shows it ahead of GPT-6 Astra on four evaluations, level on one and behind on the rest.
DeepSeek has quietly swapped the model behind its Flash API and put a price on its own search engine
The deepseek-flash endpoint now serves V4.1-Flash-1017 with no migration step. Free calls through DeepSeek Harness, Claude Code and Codex end on 1 November.
Eleven days after announcing Beam without weights, Reflection has released them. The same day, AWS made the model available through SageMaker JumpStart in two US regions.
The model is in post-training with no checkpoint, price or API announced. Microsoft plans to license the weights to partners through Foundry.
The German lab has shrunk its open image model by 38% against the version it replaces, and says it beats larger open models and two closed ones on its own evaluations.
- xAI says Grok Voice 3.0 tops the speech-to-speech index, and its own chart shows it behind on support calls
- Google says Argon leads every frontier model on its robotics reasoning test; an independent index put it third
- Nvidia's new open model activates 12 billion of its 120 billion parameters, and the small one is quantised down to 24GB
- Meta's return to open weights is an 816-billion-parameter model that it says runs on two Mac Studios
- Google has opened Gemini 4 Argon to every paying developer, and its own pricing page doubles the rate on 1 December
- Meta's first frontier model ranks fourth on an independent index, and the stronger of the pair is not sold through the API
- Odyssey has opened its new world model to the public, and its record physics score was the best of eight tries
- Perplexity's two new search models share one embedding space, so the small one can query the big one's index
- Liquid AI's 3B classifier answers in 16 milliseconds on a Jetson, and its weights are downloadable
- Cloudflare's decision model now reads audio and video, and the cheap tier fell to $0.038 with a smaller window
- Microsoft's first decision model is a post-trained Qwen, costs $0.042 a million input tokens and charges nothing for output
- OpenAI has an API that will not write you a sentence, and charges nothing for what it sends back
- Anthropic's new small model costs a tenth of the last one on short prompts, and it says the scores went up
- OpenAI is giving all 1.2 billion weekly ChatGPT users GPT-6, and it answers with interfaces rather than text
- Google halved the price of a 1K image and gave the new model a 1-million-token input window
- Mistral's new trillion-parameter model does the security work Claude and GPT-6 refuse
- Google's new embedding model puts text, images, audio and video in one space and runs in 567MB
- Amazon's new speech model keeps a 256,000-token context and calls tools without stopping the conversation
- Reka has put video, images, text and robot control into one 19-billion-parameter model
- Reflection says its new open model matches GLM 5.2 on a quarter of the inference compute