# Mistral Large 4 > Mistral Large 4 ("Le Chonk") is Mistral AI's open-weight mixture-of-experts model of about 1T parameters with 52B active, taking text and images, with three reasoning efforts and a 1M-token context, published in FP8 and NVFP4. - Canonical: https://www.anchorterminal.com/tools/mistral-large-4 - Pricing: https://www.anchorterminal.com/tools/mistral-large-4#pricing - Reviews: https://www.anchorterminal.com/tools/mistral-large-4#reviews - Markdown: https://www.anchorterminal.com/tools/mistral-large-4.md (~10,500 tokens) - Slim: https://www.anchorterminal.com/tools/mistral-large-4.min.md (~2,380 tokens, same facts, less prose, for token-sensitive contexts) - JSON: https://www.anchorterminal.com/tools/mistral-large-4.json (this page as data, same URL with Accept: application/json) - Site index for agents: https://www.anchorterminal.com/llms.txt (full text: https://www.anchorterminal.com/llms-full.txt) - API: https://www.anchorterminal.com/api/v1/index.json - MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key) - Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data) - Updated: 2026-10-30 ## Overview **Grade C · 54.6/100 · rank #92 of 117 open-weight models · #72 in Open-weight models · not agent-ready · confidence medium** More from Mistral AI, listed separately because each is its own product: [Mistral Embed and Codestral Embed](https://www.anchorterminal.com/tools/mistral-embeddings.md) (Embeddings & rerankers), [Mistral Moderation API](https://www.anchorterminal.com/tools/mistral-moderation.md) (Guardrails & safety filters), [Mistral Voxtral Transcribe](https://www.anchorterminal.com/tools/mistral-voxtral-transcribe.md) (Speech-to-text), [Voxtral Mini 4B Realtime 2602](https://www.anchorterminal.com/tools/voxtral-mini-4b-realtime-2602.md) (Speech-to-text), [Mistral OCR API](https://www.anchorterminal.com/tools/mistral-ocr.md) (Document parsing & extraction), [Mistral Vibe CLI](https://www.anchorterminal.com/tools/mistral-vibe-cli.md) (Agent harnesses). The same model, listed separately and graded on its own: [Mistral AI API](https://www.anchorterminal.com/tools/mistral-api.md) (Model API). Trained by Mistral AI, in the Mistral Large 4 family. Every model from this lab and its older versions: https://www.anchorterminal.com/open-weights/#mistral. ## Assessment A frontier-sized open cousin for coding agents and defensive security, with strict JSON Schema output, parallel tool calls and three reasoning efforts. It needs a multi-GPU server, about 732 GB for the smallest 4-bit build, and vLLM and SGLang run it only from special images. The main risk is a licence that bars Model-as-a-Service businesses without a separate deal. ## Facts | Field | Value | | --- | --- | | Vendor | Mistral AI (https://mistral.ai) | | Kind | Open-weight model | | Category | Open-weight models (https://www.anchorterminal.com/categories/open-weights) | | Auth | None · The weights download from Hugging Face with no account, token or terms to accept. Both repositories report gated false (checked 2026-10-30). | | Pricing | Free (Free) · Free to download. The Mistral Ecosystem Open-Weight License allows commercial use, except that a licensee or affiliate that operates a Model-as-a-Service business needs a separate commercial licence from Mistral before any commercial use; internal use and use through Mistral's own products are exempt. Mistral's API serves it as mistral-large-4 at $0.70 input, $0.07 cached input and $2.10 output per million tokens (checked 2026-10-30). | | x402 | No · Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-30). | | Licence | Mistral Ecosystem Open-Weight License (modified MIT) | | Packages | huggingface: `mistralai/Mistral-Large-4-1T-A52B`; huggingface: `mistralai/Mistral-Large-4-1T-A52B-NVFP4` | | Source | https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B | | Docs | https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B | | llms.txt | not found | | Last release | 2026-10-29 | | Surface graded | The open weights at huggingface.co/mistralai/Mistral-Large-4-1T-A52B (FP8) and Mistral-Large-4-1T-A52B-NVFP4, read from the model cards, the files and their metadata, the chat template, the licence, Mistral's announcements and docs, and the runtimes' own code on 30 October 2026. Nothing was downloaded or run. The hosted model is the mistral-api listing | | Model | A granular mixture of experts, 72 layers with 128 routed experts (4 active per token) and 1 shared, a dense first layer, multi-head latent attention with a sparse attention indexer, and a Pixtral-style vision encoder of 1.6B. About 1T parameters in all (1.05T in Mistral's docs) and 49B active per token, 52B with embeddings and output layers. Text and images in, text out. An embedded DSpark draft head of 3 dense layers proposes 3 tokens at a time for speculative decoding | | Files | 54 safetensors shards in block-scaled FP8 (1,060.6 GB) and 32 in the NVFP4 repository (610.0 GB, NVFP4 experts with FP8 linears), each with chat_template.jinja, params.json, tekken.json and tokenizer.json. No pickle and no Python files. Hugging Face lists a SHA-256 for every shard. No BF16 repository | | Licence | Mistral Ecosystem Open-Weight License, a modified MIT licence. Use, modification, distribution and commercial use are free, except that a licensee whose group operates a Model-as-a-Service business (giving third parties inference or post-training access with meaningful control over inputs, parameters or training data) must take a separate commercial licence before any commercial use. Internal use and use through Mistral's own products are exempt. The card adds a line forbidding use that violates third parties' rights | | Runtimes | Model code in released Transformers (mistral_large4 in v5.22.0) and llama.cpp (LLM_ARCH_MISTRAL_LARGE4 at release b12017), with GGUFs from ggml-org. vLLM and SGLang serve it only from Docker images tagged mistral-large-4; no release of either has the model. Ollama lists it as a cloud model only, and LM Studio has no entry | | Context | 1,048,576 tokens in config.json, reached with YaRN (factor 2048 over an original 512 positions) and NoPE on every fourth layer. Mistral says the checkpoint supports a 1M window and doesn't say what length it was trained at | | Tools and reasoning | The template declares tools in [AVAILABLE_TOOLS] and renders calls as [TOOL_CALLS]name[ARGS]arguments, several per turn, with results in [TOOL_RESULTS]. The card names vLLM's mistral tool and reasoning parsers and SGLang's auto parsers, and shows OpenAI-style tool calls and strict JSON Schema output through response_format, which the template passes to the model in [MODEL_SETTINGS]. reasoning_effort is low, medium or high (default) per request and can't be turned off; the reasoning comes back in message.reasoning | | Hardware | The card's commands use 8 GPUs with tensor parallelism and says NVFP4 targets Blackwell GPUs. vLLM's recipe validates GB300, B300, B200 and H200 (FP8 at about 128 GiB of weights per GPU on 8 GPUs, NVFP4 about 76 GiB) and AMD MI350X. Nothing fits a 128 GB machine; the smallest 4-bit build needs about 732 GB | | Recommended settings | temperature 1.0 and top_p 0.95 at every effort, high effort for agentic use and complex prompts, low for quick replies, a system prompt that sets the environment, and as few tools as the task needs, per the card | | Training data | Not described on the card. Mistral's preview post says it was trained from scratch on 3,800 Grace Blackwell GPUs in Mistral's own European datacentres, that a significant share of the data spans more than 160 languages, and that post-training used SFT and reinforcement learning in the same environments Mistral Forge sells to customers. No sources, sizes or cut-off. A technical report is described as coming soon | | Capabilities | model.open-weights | | Tags | open-weights, model, moe, multimodal, frontier, reasoning, tool-calling, long-context, multilingual, safetensors, fp8, nvfp4, self-hosted | | JSON | https://www.anchorterminal.com/api/v1/tools/mistral-large-4.json | ## Score breakdown (methodology v0.7, October 2026 research run) Assessed 2026-10-30 from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/#checklist). Confidence: medium. Performance and Task success pending (no score, not in the total); the total is Σ(score × weight) ÷ 80 over the 7 assessed categories. "This run" is each category's share of the 100 points. | Category | Weight | This run | Score (0–100) | Points | | --- | --- | --- | --- | --- | | Capability | 35% | 35 | 32 | 11.2 | | Reach | 25% | 25 | 50 | 12.5 | | Risk | 20% | 20 | 75 | 15.0 | | Agent interface | 10% | 10 | 95 | 9.5 | | Documentation and upkeep | 10% | 10 | 64 | 6.4 | | Negative events | up to −15 | up to −15 | none recorded | 0 | | **Total** | | | | **54.6 → C** | ### Why each score - Capability 32 (checked 2026-10-30): From 5 benchmark results (1 submitted by the publisher, at half weight) in 2 areas, each scored as its share of the way from the benchmark's floor to the best result any model has on the same leaderboard. By area, Reasoning and knowledge 34, Tool calling and agent tasks 30. Publisher-submitted results count at half weight, and numbers that appear only on the publisher's card don't count. - Reach 50 (checked 2026-10-30): Memory scores 0, since the smallest 4-bit build, Mistral's NVFP4, is 610 GB, about 732 GB with a fifth added, far above 256 GB. Runtimes score 10, released Transformers and llama.cpp have the architecture, while vLLM and SGLang run it only from special Docker images, Ollama only in its cloud and LM Studio not at all. Builds score 15 for Mistral's FP8 and NVFP4 releases. Terms score 10, a modified MIT licence that is free commercially except for Model-as-a-Service businesses. Download scores 10 with no gate, and hosted too scores 5 because Mistral's API serves these weights. - Risk 75 (checked 2026-10-30): File safety scores 25, safetensors only and loaded by Transformers' own code with a Jinja template. Provenance scores 15, Mistral's verified organisation with a hash on every file. Safety evaluation scores 10, refusal and prompt-injection results Mistral ran itself, without settings to rerun them, and red-teaming by outside firms with nothing published. Agent guidance scores 10 for a statement of what the refusals cover and a claim of trained injection resistance, with only general safeguards. Licence risk scores 10, 5 off for the Model-as-a-Service limit. The security programme scores 5 for security.txt with no public advisories. - Agent interface 95 (checked 2026-10-30): The chat template is in the repository (20). The card documents how tools are declared and come back and names the parser (25), and the template renders several calls in one turn (20). Strict JSON Schema output is documented with how to ask for it (15). Reasoning effort is budgeted per request with the reasoning returned apart from the answer (10). Context scores 5, because the 1M window comes from YaRN scaling and Mistral doesn't say it was trained at that length. - Documentation and upkeep 64 (checked 2026-10-30): The card covers intended use, architecture, parameters, context and languages, and misses limitations and a cut-off (14). Training data is described only in general terms with no cut-off (5). The files changed on 29 October 2026 (20). Mistral answered two of the five questions in the first day (10). It's a first release with older Large models in their own repositories and no dated note (5), and the files agree with each other and the card, with both reports answered (10). Fix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (12 items): https://www.anchorterminal.com/fixes/mistral-large-4.md (JSON https://www.anchorterminal.com/fixes/mistral-large-4.json) ### What we couldn't check - The trained context length isn't stated; if the promised technical report says it was trained at 1M, context moves from 5 to 10 - vLLM and SGLang support exists only in Docker images tagged mistral-large-4; runtimes rise by 5 each when a release includes the model - Whether Mistral supplied the Mistral Vibe runs on Cognition's FrontierCode 2.0 leaderboard or Cognition ran them isn't stated; recorded as Cognition's - The Artificial Analysis Cyber Index figure of 51% wasn't checked on AA in this pass - Capability isn't scored here, and confidence stays medium until the script's independent results are in ### Sources - FP8 model card, read 30 October 2026: (seen 2026-10-30) - NVFP4 model card: (seen 2026-10-30) - Hub API, gating, dates, downloads, providers and config: (seen 2026-10-30) - FP8 file tree with SHA-256 per shard, 1,060.6 GB: (seen 2026-10-30) - NVFP4 file tree, 610.0 GB: (seen 2026-10-30) - commit history of both repositories: (seen 2026-10-30) - Mistral Ecosystem Open-Weight License, read in full: (seen 2026-10-30) - chat template, tools, JSON Schema and reasoning effort: (seen 2026-10-30) - config, context and YaRN settings: (seen 2026-10-30) - discussions, authors, member flags and dates counted by script, no text read: (seen 2026-10-30) - open-weight announcement of 29 October 2026, claims, safety results, licence summary, price: (seen 2026-10-30) - preview announcement of 6 October 2026, training and red-teaming: (seen 2026-10-30) - Mistral docs model page, API names, version 26.10, weights link: (seen 2026-10-30) - security contact and HackerOne policy: (seen 2026-10-30) - llama.cpp release b12017, mistral-large4 architecture: (seen 2026-10-30) - Transformers v5.22.0 mistral_large4 model: (seen 2026-10-30) - vLLM v0.32.0 registry, no Mistral Large 4 (main checked too): (seen 2026-10-30) - vLLM recipe, no release or nightly support, validated hardware: (seen 2026-10-30) - SGLang v0.5.23, no model file (404; main checked too): (seen 2026-10-30) - Ollama library, cloud tag only: (seen 2026-10-30) - ggml-org GGUF builds and sizes: (seen 2026-10-30) - Cognition's FrontierCode leaderboard and changelog: (seen 2026-10-30) - FrontierCode 2.0 leaderboard data: (seen 2026-10-30) - Artificial Analysis model page: (seen 2026-10-30) - Artificial Analysis Coding Agent Index: (seen 2026-10-30) ## Score history Every move of a score since the listing was graded, newest first, with its cause. A score can't move without an entry here. JSON: https://www.anchorterminal.com/api/v1/tools/mistral-large-4/history.json - 2026-10-30: graded 54.6 C (methodology 0.7) ## Who's behind it (provenance 84/100, checked 2026-10-30) | Check | Finding | Points | | --- | --- | --- | | Legal entity named | Mistral AI (RCS Paris 952 418 325) | 20/20 | | Domain age | mistral.ai, registered 2019-05-15 (7 years) | 11/15 | | Endpoint on the vendor's domain | no hosted endpoint | n/a | | Terms of service | nothing hosted, so the Mistral Ecosystem Open-Weight License (modified MIT) licence stands in | 10/10 | | Privacy policy | nothing hosted, not scored | n/a | | Status page | not found | 0/10 | | Changelog | published | 10/10 | | security.txt | valid | 10/10 | | People named | 3 named with their role, 3 with a public profile | 10/10 | | Support route | help pages, a community | 5/5 | The files sit in Mistral's verified mistralai organisation, which the card, Mistral's announcement and its docs page for the model link to. Legal entity and domain registration are carried from the mistral-api listing. Nothing is hosted in this listing, so the licence stands in for terms of service and no privacy policy is scored. The hosted model is mistral-api. The changelog is the repositories' commit history, an initial release on 29 October 2026 and two card edits the same day, plus a merged evaluation figure on the NVFP4 card. mistral.ai/.well-known/security.txt answered on 30 October 2026 with a security contact, an expiry of 5 May 2027 and a HackerOne submission policy. ### Terms and privacy, as read A reading by a fixed set of rules, each answered with the vendor's own sentence. Not legal advice. **Terms of service**. Nothing is hosted by the vendor, so there are no terms of service to read. The Mistral Ecosystem Open-Weight License (modified MIT) licence stands in and the check scores in full. **Privacy policy**. Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored. ## Humans behind the project (researched 2026-10-30 from public pages, not part of the grade) ### Key people - Arthur Mensch, Co-founder and CEO. LinkedIn: https://www.linkedin.com/in/arthur-mensch/ - Guillaume Lample, Co-founder and Chief Science Officer. LinkedIn: https://www.linkedin.com/in/guillaume-lample-7821095b/ - Timothee Lacroix, Co-founder and CTO. LinkedIn: https://www.linkedin.com/in/timothee-lacroix-59517977/ - Note: the three founders and their roles are carried from data/humans/mistral-api.json (checked 28 october 2026, from mistral.ai/about). mistral's announcement and model card for large 4 are signed by mistral and name no product leads; the hugging face repositories' commits are by mistral staff accounts, which aren't listed here as roles. ownership and investors are left out so land.py carries them over from the mistral-api humans file (checked 28 october 2026: private, series d of 3 billion euros at more than 21 billion euros post-money in september 2026); nothing newer was looked for two days later. - Mistral AI: LinkedIn https://www.linkedin.com/company/mistralai/ X https://x.com/mistralai GitHub https://github.com/mistralai ### Ownership and investors - Valuation: more than €21B post-money, Series D, September 2026 (source https://mistral.ai/news/mistral-makes-sovereign-open-weight-ai-to-frontier/) - Backed by: Lightspeed Venture Partners, JCDecaux Holding, Exor, Sofina, Eric Schmidt, Xavier Niel, Rodolphe Saadé, Andreessen Horowitz, General Catalyst, NVIDIA, Samsung Venture Investment Corporation, Salesforce, Belfius, Bertelsmann, BNP Paribas, Bpifrance, Cisco Investments, Eurazeo, Headline, Hanwha Asset Management, IBM, Korelya Capital, Latitude Capital, Millennium New Horizons, Sanabil Investments, ServiceNow, SV Angel, Microsoft, ASML Holding, DST Global, Index Ventures, Samsung Electronics, EQT, PSG Equity, Advent, BlackRock, Grand Duchy of Luxembourg, Carmignac, Hillspire, Phoenix Court ### Getting support - Discord: https://discord.gg/mistralai - Help: https://help.mistral.ai/en/articles/347458-how-do-i-contact-support - Help: https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/discussions - Community (Reddit): https://www.reddit.com/r/MistralAI/ - Sources: https://mistral.ai/about/, https://www.linkedin.com/in/arthur-mensch/, https://www.linkedin.com/in/guillaume-lample-7821095b/, https://www.linkedin.com/in/timothee-lacroix-59517977/, https://mistral.ai/, https://help.mistral.ai/en/articles/347458-how-do-i-contact-support, https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B, https://mistral.ai/news/mistral-large-4-oss/ ## Probe metrics Not measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. Live uptime, where we poll the endpoint, is under Live and doesn't change the score. ## What a task costs hosted (estimated, per task) | Hosted as | Research report | Small website | Bug fix | | --- | --- | --- | --- | | `mistralai/mistral-large-4-0` (OpenRouter, $0.70 in and $2.10 out per 1M tokens) | $0.98 | $0.31 | $0.08 | Each figure is the same tokens priced at this model's rates, research report (134 fresh input, 207k cache writes, 10.5M cache reads and 46k output, the median of 156 runs), small website (48 fresh input, 106k cache writes, 2.2M cache reads and 40k output, the median of 4 runs) and bug fix (16 fresh input, 57k cache writes, 472k cache reads and 2k output, the median of 4 runs). We measured the tokens with Claude Opus 5.5 in Claude Code, so another model may use more or fewer tokens for the same job, and these compare prices, not how well or how efficiently each model works. Long-context surcharges, batch discounts and free allowances aren't counted. Prices are OpenRouter's for each model it serves from these weights. Run them yourself and the cost is your hardware and power. How we measured them: https://www.anchorterminal.com/prices/index.md ## Strengths - Safetensors only, loaded by Transformers' own code with a Jinja template, in Mistral's verified organisation with a hash on every file - Tool calling documented end to end with a named parser, several calls a turn and strict JSON Schema output - Reasoning effort budgeted per request, low, medium or high, with the reasoning returned apart from the answer - FP8 and NVFP4 builds from Mistral, and Mistral's own API serves the same model for agents without the hardware - First among open-weight models on Cognition's FrontierCode 2.0 leaderboard, run in Mistral's own harness ## Weaknesses - The smallest 4-bit build needs about 732 GB, so nothing fits a single workstation or a 128 GB machine - No vLLM or SGLang release supports it yet, only Docker images tagged mistral-large-4 - A Model-as-a-Service business needs a separate commercial licence from Mistral before any commercial use - Safety results are Mistral's own with no settings to rerun them, and no limitations, training data sources or knowledge cut-off are published - The 1M context comes from YaRN scaling with no stated trained length, and thinking can't be turned off ## Before you call it (notes for agents) 1. Serve the NVFP4 build on Blackwell with `vllm/vllm-openai:mistral-large-4 --model mistralai/Mistral-Large-4-1T-A52B-NVFP4 --tensor-parallel-size 8 --kv-cache-dtype fp8_ds_mla --tool-call-parser mistral --reasoning-parser mistral --enable-auto-tool-choice`; released vLLM doesn't load it 2. On H200 use the FP8 build with tensor parallelism of 16 or 8 by 2 pipelines, per vLLM's recipe 3. Pass `reasoning_effort` low, medium or high (default high); any other value makes the template raise an error, and there is no off switch 4. Read the reasoning from `message.reasoning`; `max_completion_tokens` counts reasoning tokens too 5. Use temperature 1.0 and top_p 0.95 at every effort, per the card 6. Ask for JSON with `response_format` type json_schema and strict true; the template passes the schema to the model 7. The template rejects a tool message straight after a user message and a conversation that doesn't start with a user or system message 8. Add `--speculative-config '{"method": "dspark", "num_speculative_tokens": 3}'` to use the embedded draft head 9. If you'll offer it to others as a hosted model, read the licence first; Model-as-a-Service use needs Mistral's commercial licence ## Get started Install: ```bash # NVFP4 on 4 or 8 Blackwell GPUs, with the card's vLLM image (no vLLM release supports it yet): docker run --runtime nvidia --gpus all -v ~/.cache/huggingface:/root/.cache/huggingface -p 8000:8000 --ipc=host \ vllm/vllm-openai:mistral-large-4 --model mistralai/Mistral-Large-4-1T-A52B-NVFP4 \ --tensor-parallel-size 8 --kv-cache-dtype fp8_ds_mla --tool-call-parser mistral \ --reasoning-parser mistral --enable-auto-tool-choice --gpu-memory-utilization 0.85 ``` First request: ```bash curl http://localhost:8000/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model": "mistralai/Mistral-Large-4-1T-A52B-NVFP4", "reasoning_effort": "medium", "temperature": 1.0, "top_p": 0.95, "messages": [{"role": "user", "content": "Hello!"}]}' ``` ## Similar tools Ranked by shared capabilities, then score. Same-category tools with no shared capability key are listed last. | Tool | Grade | Score | Rank | Shared capabilities | x402 | Markdown | | --- | --- | --- | --- | --- | --- | --- | | Qwen3-Embedding-8B | BB | 74.7 | open-weight #1 | model.open-weights | no | https://www.anchorterminal.com/tools/qwen3-embedding-8b.md | | Gemma 4 31B IT | BB | 74.1 | open-weight #2 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-31b-it.md | | Qwen3-Embedding-4B | BB | 72.9 | open-weight #3 | model.open-weights | no | https://www.anchorterminal.com/tools/qwen3-embedding-4b.md | | Gemma 4 26B A4B IT | BB | 72.2 | open-weight #4 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.md | | Gemma 4 12B IT | BB | 71.3 | open-weight #5 | model.open-weights | no | https://www.anchorterminal.com/tools/gemma-4-12b-it.md | | gpt-oss-120b | BB | 70.2 | open-weight #6 | model.open-weights | no | https://www.anchorterminal.com/tools/gpt-oss-120b.md | ## Panel reviews (0) Reviewed by the Anchor panel (https://www.anchorterminal.com/reviewers/index.md): . Desk reviews, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. For a desk review, the outcome says whether the reviewer's questions could be answered from public material: success, partial or failure. How reviews work: https://www.anchorterminal.com/reviews/how-it-works.md ## Notable - Released as open weights on 29 October 2026 after a public preview on Mistral's API from 6 October. Mistral says three more weeks of reinforcement learning went into the released version (source: ) - 1,239 downloads and 1,329 likes for the FP8 repository and 315 downloads for NVFP4 on Hugging Face on 30 October 2026, a day after release (source: ) - Cognition's FrontierCode 2.0 leaderboard puts it first of seven open-weight models at 19.65%, run in Mistral's own Vibe harness, against DeepSeek V4.1 Flash at 19.3% in devin-cli. Artificial Analysis's own run of the same benchmark gives 15.0%, 10th of 31 open-weight rows (source: ) - Artificial Analysis gives it 26.0 on its Intelligence Index v5.0, 9th of 31 open-weight models and the highest from a western lab, and 48.3% on its Coding Agent Index in Mistral Vibe, 0.6 points behind GLM-5.3 in Opencode with 47% fewer tokens per task (source: ) - Mistral reports a 96.6% refusal rate on malicious cyber prompts from JailbreakBench, StrongREJECT and AgentHarm, and 97.1% attack resistance on Lakera's B3 agent security benchmark. These are Mistral's own numbers (source: ) - The announcement quotes Booz Allen, Cisco, Cloudflare, CrowdStrike, Elastic, ENISA, Fortinet, Orange Cyberdefense, Rubrik and Zscaler on cyber defence testing, with no results published, and says customers tailor it with Mistral Forge, Mistral's customisation service (source: ) - No vLLM or SGLang release supports it yet. The card's commands use Docker images tagged mistral-large-4, and vLLM's recipe says no release or nightly wheel includes it (source: ) - #72 of 85 in Best open-weight models you download and run yourself for AI agents: https://www.anchorterminal.com/best/open-weights/index.md - All 3588 open weights comparisons: https://www.anchorterminal.com/compare/open-weights/index.md ## Compare - [Agents-A1-4B vs Mistral Large 4](https://www.anchorterminal.com/compare/agents-a1-4b-vs-mistral-large-4.md): C 61 vs C 54.6 - [Apertus v1.5 8B vs Mistral Large 4](https://www.anchorterminal.com/compare/apertus-v1-5-8b-vs-mistral-large-4.md): C 61 vs C 54.6 - [Apodex-1.1-mini vs Mistral Large 4](https://www.anchorterminal.com/compare/apodex-1-1-mini-vs-mistral-large-4.md): C 61 vs C 54.6 - [BigBang-v1 vs Mistral Large 4](https://www.anchorterminal.com/compare/bigbang-v1-vs-mistral-large-4.md): C 54.4 vs C 54.6 - [Carbon-X vs Mistral Large 4](https://www.anchorterminal.com/compare/carbon-x-vs-mistral-large-4.md): C 61 vs C 54.6 - [Cosmos-Reason2-2B vs Mistral Large 4](https://www.anchorterminal.com/compare/cosmos-reason2-2b-vs-mistral-large-4.md): C 59.5 vs C 54.6 - [Darwin-36B-Opus vs Mistral Large 4](https://www.anchorterminal.com/compare/darwin-36b-opus-vs-mistral-large-4.md): C 61 vs C 54.6 - [DeepSeek-V3.2 vs Mistral Large 4](https://www.anchorterminal.com/compare/deepseek-v3-2-vs-mistral-large-4.md): C 58.6 vs C 54.6 - [DeepSeek V4.1 Flash vs Mistral Large 4](https://www.anchorterminal.com/compare/deepseek-v4-1-flash-vs-mistral-large-4.md): C 59 vs C 54.6 - [DeepSeek-V4-Flash-0731 vs Mistral Large 4](https://www.anchorterminal.com/compare/deepseek-v4-flash-0731-vs-mistral-large-4.md): B 62.4 vs C 54.6 - [DiffusionGemma 26B A4B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/diffusiongemma-26b-a4b-it-vs-mistral-large-4.md): D 50 vs C 54.6 - [Gemma 4 12B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/gemma-4-12b-it-vs-mistral-large-4.md): BB 71.3 vs C 54.6 - [Gemma 4 26B A4B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/gemma-4-26b-a4b-it-vs-mistral-large-4.md): BB 72.2 vs C 54.6 - [Gemma 4 31B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/gemma-4-31b-it-vs-mistral-large-4.md): BB 74.1 vs C 54.6 - [Gemma 4 E2B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/gemma-4-e2b-it-vs-mistral-large-4.md): C 60.4 vs C 54.6 - [Gemma 4 E4B IT vs Mistral Large 4](https://www.anchorterminal.com/compare/gemma-4-e4b-it-vs-mistral-large-4.md): B 63.2 vs C 54.6 - [GLM-4.5-Air vs Mistral Large 4](https://www.anchorterminal.com/compare/glm-4-5-air-vs-mistral-large-4.md): C 59.1 vs C 54.6 - [GLM-4.7-Flash vs Mistral Large 4](https://www.anchorterminal.com/compare/glm-4-7-flash-vs-mistral-large-4.md): B 63 vs C 54.6 - [GLM-5.3-Flash vs Mistral Large 4](https://www.anchorterminal.com/compare/glm-5-3-flash-vs-mistral-large-4.md): B 65.1 vs C 54.6 - [GLM-5.3 vs Mistral Large 4](https://www.anchorterminal.com/compare/glm-5-3-vs-mistral-large-4.md): B 64.6 vs C 54.6 - [gpt-oss-120b vs Mistral Large 4](https://www.anchorterminal.com/compare/gpt-oss-120b-vs-mistral-large-4.md): BB 70.2 vs C 54.6 - [gpt-oss-20b vs Mistral Large 4](https://www.anchorterminal.com/compare/gpt-oss-20b-vs-mistral-large-4.md): B 69.9 vs C 54.6 - [Granite 4.2 3B vs Mistral Large 4](https://www.anchorterminal.com/compare/granite-4-2-3b-vs-mistral-large-4.md): C 59.8 vs C 54.6 - [Hy3 vs Mistral Large 4](https://www.anchorterminal.com/compare/hy3-vs-mistral-large-4.md): B 62.6 vs C 54.6 - [Inkling-Small vs Mistral Large 4](https://www.anchorterminal.com/compare/inkling-small-vs-mistral-large-4.md): B 66.5 vs C 54.6 - [Inkling vs Mistral Large 4](https://www.anchorterminal.com/compare/inkling-vs-mistral-large-4.md): B 64.2 vs C 54.6 - [InternVL3.5-GPT-OSS-20B-A4B-Preview vs Mistral Large 4](https://www.anchorterminal.com/compare/internvl3-5-gpt-oss-20b-a4b-preview-vs-mistral-large-4.md): D 51.2 vs C 54.6 - [Kimi K3 vs Mistral Large 4](https://www.anchorterminal.com/compare/kimi-k3-vs-mistral-large-4.md): B 62.4 vs C 54.6 - [Kolibri-1 vs Mistral Large 4](https://www.anchorterminal.com/compare/kolibri-1-vs-mistral-large-4.md): C 61 vs C 54.6 - [Kronos vs Mistral Large 4](https://www.anchorterminal.com/compare/kronos-vs-mistral-large-4.md): C 55.2 vs C 54.6 - [Laguna M.1 vs Mistral Large 4](https://www.anchorterminal.com/compare/laguna-m-1-vs-mistral-large-4.md): C 55.5 vs C 54.6 - [Laguna S 2.1 vs Mistral Large 4](https://www.anchorterminal.com/compare/laguna-s-2-1-vs-mistral-large-4.md): C 61 vs C 54.6 - [Laguna XS 2.1 vs Mistral Large 4](https://www.anchorterminal.com/compare/laguna-xs-2-1-vs-mistral-large-4.md): C 61 vs C 54.6 - [LBS-27B vs Mistral Large 4](https://www.anchorterminal.com/compare/lbs-27b-vs-mistral-large-4.md): D 53.9 vs C 54.6 - [LFM2.5-2.6B vs Mistral Large 4](https://www.anchorterminal.com/compare/lfm2-5-2-6b-vs-mistral-large-4.md): D 51.5 vs C 54.6 - [Ling-3.0-tiny vs Mistral Large 4](https://www.anchorterminal.com/compare/ling-3-0-tiny-vs-mistral-large-4.md): C 55 vs C 54.6 - [MiMo-V2.6 Distill Qwen 9B vs Mistral Large 4](https://www.anchorterminal.com/compare/mimo-v2-6-distill-qwen-9b-vs-mistral-large-4.md): C 61 vs C 54.6 - [MiniCPM-o 4.5 vs Mistral Large 4](https://www.anchorterminal.com/compare/minicpm-o-4-5-vs-mistral-large-4.md): C 61 vs C 54.6 - [MiniCPM5-2B vs Mistral Large 4](https://www.anchorterminal.com/compare/minicpm5-2b-vs-mistral-large-4.md): C 60.2 vs C 54.6 - [MiniMax-M2.5 vs Mistral Large 4](https://www.anchorterminal.com/compare/minimax-m2-5-vs-mistral-large-4.md): C 57.5 vs C 54.6 - [MiniMax-M2.7 vs Mistral Large 4](https://www.anchorterminal.com/compare/minimax-m2-7-vs-mistral-large-4.md): C 56.1 vs C 54.6 - [MiniMax-M3.1-Flash vs Mistral Large 4](https://www.anchorterminal.com/compare/minimax-m3-1-flash-vs-mistral-large-4.md): D 50.5 vs C 54.6 - [Mistral Large 4 vs Muse Glimmer 30B](https://www.anchorterminal.com/compare/mistral-large-4-vs-muse-glimmer-30b.md): C 54.6 vs BB 70 - [Mistral Large 4 vs Nanbeige4.3-33B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nanbeige-4-3-33b.md): C 54.6 vs C 61 - [Mistral Large 4 vs NVIDIA Nemotron 3.5 Bolt E4B 4GB](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-5-bolt-e4b-4gb.md): C 54.6 vs C 61 - [Mistral Large 4 vs Nemotron 3.5 Lightning 30B A3B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-5-lightning-30b-a3b.md): C 54.6 vs B 64.2 - [Mistral Large 4 vs Nemotron 3 Nano 30B A3B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-nano-30b-a3b.md): C 54.6 vs B 67.1 - [Mistral Large 4 vs Nemotron 3 Nano 4B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-nano-4b.md): C 54.6 vs C 56.8 - [Mistral Large 4 vs Nemotron 3 Nano Omni 30B A3B Reasoning](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-nano-omni-30b-a3b-reasoning.md): C 54.6 vs C 60.7 - [Mistral Large 4 vs Nemotron 3 Super 120B A12B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-super-120b-a12b.md): C 54.6 vs B 66.2 - [Mistral Large 4 vs Nemotron 3 Ultra 550B A55B](https://www.anchorterminal.com/compare/mistral-large-4-vs-nemotron-3-ultra-550b-a55b.md): C 54.6 vs B 68.8 - [Mistral Large 4 vs occamy-1.0](https://www.anchorterminal.com/compare/mistral-large-4-vs-occamy-1-0.md): C 54.6 vs C 61 - [Mistral Large 4 vs Olmo 3 7B Think](https://www.anchorterminal.com/compare/mistral-large-4-vs-olmo-3-7b-think.md): C 54.6 vs D 52.3 - [Mistral Large 4 vs Ornith-1.5-35B-A3B](https://www.anchorterminal.com/compare/mistral-large-4-vs-ornith-1-5-35b-a3b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Ornith-1.5-397B](https://www.anchorterminal.com/compare/mistral-large-4-vs-ornith-1-5-397b.md): C 54.6 vs C 59 - [Mistral Large 4 vs Ornith-1.5-9B](https://www.anchorterminal.com/compare/mistral-large-4-vs-ornith-1-5-9b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Qwen3-30B-A3B-Instruct-2507](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-30b-a3b-instruct-2507.md): C 54.6 vs C 56.6 - [Mistral Large 4 vs Qwen3-4B-Instruct-2507](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-4b-instruct-2507.md): C 54.6 vs C 55.7 - [Mistral Large 4 vs Qwen3.5-0.8B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-0-8b.md): C 54.6 vs D 46.7 - [Mistral Large 4 vs Qwen3.5-122B-A10B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-122b-a10b.md): C 54.6 vs B 63.6 - [Mistral Large 4 vs Qwen3.5-2B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-2b.md): C 54.6 vs D 50.5 - [Mistral Large 4 vs Qwen3.5-397B-A17B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-397b-a17b.md): C 54.6 vs B 64.6 - [Mistral Large 4 vs Qwen3.5-4B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-4b.md): C 54.6 vs C 61.2 - [Mistral Large 4 vs Qwen3.5-9B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-5-9b.md): C 54.6 vs C 60.5 - [Mistral Large 4 vs Qwen3.6-27B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-6-27b.md): C 54.6 vs B 67.2 - [Mistral Large 4 vs Qwen3.6-35B-A3B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-6-35b-a3b.md): C 54.6 vs B 63.2 - [Mistral Large 4 vs Qwen3.8-27B](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-8-27b.md): C 54.6 vs B 68.6 - [Mistral Large 4 vs Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-8-flash-next.md): C 54.6 vs C 61.8 - [Mistral Large 4 vs Qwen3-Coder-30B-A3B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-coder-30b-a3b-instruct.md): C 54.6 vs C 55.7 - [Mistral Large 4 vs Qwen3-Coder-Next](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-coder-next.md): C 54.6 vs C 57.6 - [Mistral Large 4 vs Qwen3-Omni-30B-A3B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-omni-30b-a3b-instruct.md): C 54.6 vs E 41.8 - [Mistral Large 4 vs Qwen3-VL-2B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-vl-2b-instruct.md): C 54.6 vs C 61 - [Mistral Large 4 vs Qwen3-VL-30B-A3B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-vl-30b-a3b-instruct.md): C 54.6 vs C 58 - [Mistral Large 4 vs Qwen3-VL-32B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-vl-32b-instruct.md): C 54.6 vs C 56.4 - [Mistral Large 4 vs Qwen3-VL-4B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-vl-4b-instruct.md): C 54.6 vs D 52.6 - [Mistral Large 4 vs Qwen3-VL-8B-Instruct](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwen3-vl-8b-instruct.md): C 54.6 vs C 54.9 - [Mistral Large 4 vs Qwopus3.8-27B-Flash-V2](https://www.anchorterminal.com/compare/mistral-large-4-vs-qwopus3-8-27b-flash-v2.md): C 54.6 vs C 61 - [Mistral Large 4 vs Recursive Flint](https://www.anchorterminal.com/compare/mistral-large-4-vs-recursive-flint.md): C 54.6 vs C 61 - [Mistral Large 4 vs SmolLM3-3B](https://www.anchorterminal.com/compare/mistral-large-4-vs-smollm3-3b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Spark-X2.6-4B](https://www.anchorterminal.com/compare/mistral-large-4-vs-spark-x2-6-4b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Swift 1.5 Qwen3.8-27B](https://www.anchorterminal.com/compare/mistral-large-4-vs-swift-1-5-qwen3-8-27b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Swift 1.5 Qwen3.8-Flash-Next](https://www.anchorterminal.com/compare/mistral-large-4-vs-swift-1-5-qwen3-8-flash-next.md): C 54.6 vs D 52.3 - [Mistral Large 4 vs ThinkingCap (Qwen3.8 27B)](https://www.anchorterminal.com/compare/mistral-large-4-vs-thinkingcap-qwen3-8-27b.md): C 54.6 vs C 61 - [Mistral Large 4 vs Baidu Unlimited-OCR-2](https://www.anchorterminal.com/compare/mistral-large-4-vs-unlimited-ocr-2.md): C 54.6 vs D 53.5 ## Verify this listing For the vendor. The badge or a plain link to this page verifies the listing, from a page on mistral.ai or one of its subdomains. It shows the listing is the vendor's and that the vendor knows it's here, and it never changes a grade, rank or review. The vendor sends the page's address to `POST https://www.anchorterminal.com/api/v1/verify` as `{"slug": "mistral-large-4", "url": "…"}`, or calls the `verify_listing` tool at https://www.anchorterminal.com/mcp. We fetch the page once, then again every week; two failed checks in a row and the verification lapses, and a later pass restores it. What we check: https://www.anchorterminal.com/builders/index.md#verify HTML badge: ```html Mistral Large 4 on Anchor Terminal ``` Markdown badge, for a README: ```markdown [![Mistral Large 4 on Anchor Terminal](https://www.anchorterminal.com/badges/mistral-large-4.svg)](https://www.anchorterminal.com/tools/mistral-large-4) ``` Plain link: ```html Mistral Large 4 on Anchor Terminal ``` ## Share this listing For the vendor. Sharing assets for social media, two PNGs of 1200 × 630 that say Mistral Large 4 is listed on Anchor Terminal, with the vendor's logo and this page's address and no grade or score. - Dark: https://www.anchorterminal.com/assets/share/mistral-large-4-dark.png - Light: https://www.anchorterminal.com/assets/share/mistral-large-4-light.png - Create a video and download its frames: https://www.anchorterminal.com/assets/listing-video/index.html?listing=mistral-large-4 --- Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp