Mistral Large 4 1T

Mistral Large 4 1T MoE (52B active) on NVIDIA NIM with NVFP4 and DSpark speculative decoding — 40 RPM.

Updated 11/22/20261 code examples

About the Model

为什么要选择 Mistral Large 4?

Mistral Large 4 是 Mistral AI 联合 NVIDIA 打造的 1 万亿(1T)参数细粒度 MoE 旗舰大模型(激活参数 52B)。具备卓越的图文多模态推理、精准工具调用(Function Calling)与结构化 JSON 输出能力,以 NVFP4 精度提供超高吞吐服务。

核心优势与技术突破

  • 1T 细粒度 MoE 架构:52B 动态激活,在代码重构、数学论证与多语言推理中对齐世界顶级闭源水平。
  • DSpark 推测解码:结合 NVIDIA TensorRT-LLM 投机采样算法,Token 生成速度翻倍提升。
  • 原生多模态与严谨结构化输出:支持图文混合输入,在遵循复杂 JSON Schema 与工具编排时稳定性极高。

How to Access for Free (via NVIDIA NIM)

Why Choose Mistral Large 4 on NVIDIA NIM?

Mistral Large 4 (1T-total / 52B-active) is served on NVIDIA NIM in NVFP4 with DSpark speculative decoding. Delivering frontier performance in complex programming, mathematical logic, and function calling with zero infrastructure overhead.

Try it in your browser

Pick a model, paste your own free key, and run. Your key is sent once to call the provider and never stored on our servers.

Pick a model, paste your own free key, and run. Your key is sent once to call the provider and never stored on our servers.

Code Examples

curl
curl https://integrate.api.nvidia.com/v1/chat/completions \
  -H "Authorization: Bearer $NVIDIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistralai/mistral-large-4-1t-a52b",
    "messages": [{"role": "user", "content": "Explain lock-free concurrent programming algorithms in C++20."}]
  }'

Community Feedback

Comments are tied to your GitHub account — sign in to join the discussion.