Mistral Large 4
A flagship open-source large language model released by Mistral, built on a Mixture-of-Experts (MoE) architecture. It has approximately one trillion total parameters, with roughly 50 billion parameters activated per token. The model supports multimodal input and an extended context window of up to one million tokens.
Core Facts
Mistral Large 4 is the largest open-source model Mistral has released to date
Mistral Large 4 was trained using 3,800 NVIDIA Grace Blackwell GPUs in its own data centers, and the public preview also runs on the same infrastructure.
Mistral Large 4 uses a MoE architecture with 1 trillion total parameters and 49 billion active parameters, is natively multimodal, and has a context window of 512,000 tokens.
Timeline (last 90 days)
The updated Mistral Large 4 weights focus on enhancing agentic coding and defensive cybersecurity capabilities.
To use Mistral Large 4 in coding agent tools such as fx, Pi, Claude Code, Codex, and Cursor, you need to install the latest Vercel CLI and run `npx vercel ai-gateway setup`.
The prices for Mistral Large 4's input tokens, cached input tokens, and output tokens have all been lowered.
The code path for calling Mistral Large 4 on AI Gateway has not changed; when using the AI SDK, you still pass `mistral/mistral-large-4` and call `streamText`.
The updated Mistral Large 4 weights have higher token efficiency in coding tasks, are more accurate at table creation and editing, have enhanced web research capabilities, and have improved resistance to prompt injection.
Mistral Large 4 updated its model weights on AI Gateway and lowered its price at the same time, while the model ID `mistral/mistral-large-4` remains unchanged.
Mistral embedded extensive refusal behavior in ML4, and Pierre said its refusal scope far exceeds that of Chinese competitors.
Mistral brought its first batch of 4K chips on its own compute online this year, and Le Chanc was trained on this infrastructure.
The model is now available on Hugging Face, and the blog provides access methods via Transformers, llama.cpp, vLLM, SGLang, as well as Mistral's own platform and its coding agent Le Vibe.
ML4 has a built-in length penalty mechanism and is 50% more efficient than GLM 5.3 on the coding index.
40 more timeline events
All Facts (20)
Mistral Large 4 is the largest open-source model Mistral has released to date
75%VerifiedMistral Large 4 was trained using 3,800 NVIDIA Grace Blackwell GPUs in its own data centers, and the public preview also runs on the same infrastructure.
70%VerifiedMistral Large 4 uses a MoE architecture with 1 trillion total parameters and 49 billion active parameters, is natively multimodal, and has a context window of 512,000 tokens.
65%VerifiedMistral Large 4 narrowly outperformed GPT-6 Astra on visual grounding tasks with a score of 42% versus 41%.
65%VerifiedMistral Large 4 is a hybrid instruction-and-reasoning MoE model with a total of 1 trillion parameters, activating only approximately 49 billion parameters during inference.
65%VerifiedMistral has released Mistral Large 4, a one-trillion-parameter, natively multimodal MoE model with approximately 49 billion active parameters, with open-source weights to be released at the end of the month.
75%VerifiedMistral released Mistral Large 4, codenamed LeChunk, a trillion-parameter, natively multimodal MoE model with approximately 49 billion active parameters.
85%VerifiedMistral has launched its new flagship model Mistral Large 4, codenamed 'Le Chonk'.
80%UnverifiedMistral Medium 3.5是Mistral于2025年4月发布的旗舰模型，专为智能体和编码工作场景优化
90%Partially VerifiedMistral Large 4 adopts a sparse Mixture of Experts (MoE) architecture, with 49 billion (49B) parameters actually activated during each inference.
75%UnverifiedMistral Large 4 was trained entirely from scratch in Mistral's own European data centers using 3,800 Grace Blackwell GPUs, with no distillation performed.
60%UnverifiedFrance's Mistral has released Mistral Large 4, scoring 38 points in a third-party intelligence index evaluation.
60%UnverifiedThe code path for calling Mistral Large 4 on AI Gateway has not changed; when using the AI SDK, you still pass `mistral/mistral-large-4` and call `streamText`.
50%UnverifiedTo use Mistral Large 4 in coding agent tools such as fx, Pi, Claude Code, Codex, and Cursor, you need to install the latest Vercel CLI and run `npx vercel ai-gateway setup`.
50%UnverifiedThe updated Mistral Large 4 weights have higher token efficiency in coding tasks, are more accurate at table creation and editing, have enhanced web research capabilities, and have improved resistance to prompt injection.
50%UnverifiedThe prices for Mistral Large 4's input tokens, cached input tokens, and output tokens have all been lowered.
50%UnverifiedThe updated Mistral Large 4 weights focus on enhancing agentic coding and defensive cybersecurity capabilities.
50%UnverifiedOn Artificial Analysis's new coding index, ML4 scores the same as GLM 5.3, ahead of other Western open-source models.
50%UnverifiedML4 has a built-in length penalty mechanism and is 50% more efficient than GLM 5.3 on the coding index.
50%UnverifiedMistral embedded extensive refusal behavior in ML4, and Pierre said its refusal scope far exceeds that of Chinese competitors.
50%