Mistral-Large-4

Name in the API

mistralai/Mistral-Large-4

Base model

mistralai/Mistral-Large-4-1T-A52B

Weights we run

mistralai/Mistral-Large-4-1T-A52B-NVFP4 (NVFP4 quantization: experts in 4 bit, attention and shared experts in FP8)

Type

Chat

Input → output

text, image → text

Context length

524,288 tokens

Availability

HPC

Hardware

8 × NVIDIA H100 (94 GB) on two nodes of the Capella HPC cluster

Thinking

Default: on (high).

reasoning_effort: low, medium, high. Rejected with error 400: none, minimal, xhigh, max.

Numeric effort: no.

Turn off: not possible. low keeps the reasoning short: for a one-sentence answer about 90 tokens instead of about 400.

chat_template_kwargs is rejected with error 400 for this model.

Tool calling

Yes

Limits

At most 4 images per request. Video input is not supported.

At most 65,536 output tokens per response.

Example

Chat request. Set model to the name above.

Notes

• Long prompts take time on first read: about 30 seconds for 130K tokens and about 3 minutes for 500K tokens. Repeated prefixes, such as follow-up turns or the same document, are cached and answered much faster.
• Mistral recommends temperature: 1.0 for all reasoning efforts.
• License: Mistral Ecosystem Open-Weight License.