As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/m/mistral-large-4/ # Mistral Large 4 Mistral AI, Mistral Large family. Status: current. Type: multimodal model. Released: 2026-10-06. Training cutoff: not published. ## How to call it | Provider | Model id | Endpoint | Docs | |---|---|---|---| | Mistral API | `mistral-large-4-0` | https://api.mistral.ai/v1/chat/completions | https://docs.mistral.ai/models/mistral-large-4-0 | | Hugging Face | | | https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B | | Hugging Face (NVFP4) | | | https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B-NVFP4 | | OpenRouter | `mistralai/mistral-large-4-0` | | | | NVIDIA (build.nvidia.com) | `mistralai/mistral-large-4-1t-a52b` | | | | Microsoft Foundry | | | https://ai.azure.com/ | | Web app | | | https://chat.mistral.ai | ## Pricing - Input: $0.70 per 1M tokens - Cached input: $0.07 per 1M tokens - Output: $2.10 per 1M tokens - Unit as published: per 1M tokens (USD), Mistral API, from the Oct 29, 2026 weights release - Source: https://mistral.ai/news/mistral-large-4-oss/ - Last checked: 2026-10-29 ## Specs - Input: text, image - Output: text - Context window: 1,000,000 tokens - Max output: not published - Open weights: yes - Licence: Mistral Ecosystem Open-Weight License (modified MIT; model-as-a-service businesses need a commercial licence) ## What it has not seen - No published training cutoff. Counted from its release (2026-10-06): at least 1,143 AI events since (230 major, 25 historic), as of 2026-11-22. - Everything since September 2026: https://postcutoff.com/since/2026-09/ ## What stands out - Trillion-parameter open-weight European model (first): Granular MoE, ~1T total / 49B active parameters (52B with embeddings), 1M context, 1.6B vision encoder; weights released Oct 29, 2026 in FP8 and NVFP4. Source: https://huggingface.co/blog/mistral-large4 - Cyber defence: Mistral reports 51% on the Artificial Analysis Cyber Index and 96.6% refusal of malicious cyber requests; a specialised cyber variant is gated to vetted partners. Source: https://mistral.ai/news/mistral-large-4-oss/ - Three reasoning levels: Low, Medium and High (default); reasoning cannot be switched off in the released checkpoint. Source: https://huggingface.co/blog/mistral-large4 ## Notes Preview from Oct 6, 2026 (entry 2026-10-06-mistral-large-4); further-trained weights and price cut on Oct 29. Artificial Analysis Intelligence Index v5: 26 (High). Self-hosting: NVFP4 needs Blackwell GPUs; 2-bit GGUF fits a 512 GB Apple Silicon Ultra. Mistral's flagship open-weight MoE, successor to Mistral Large 3 (Apache 2.0). The Oct 29 release moved to a custom licence. Self-reported: Terminal-Bench 4.0 41%, AA Coding Agent Index 48% (about level with GLM-5.3), Cognition FrontierCode 2.0 19.7%. Model page: https://postcutoff.com/m/mistral-large-4/