Mistral Small 4
Mistral AI · Multimodal model
Mistral Small 4 is an efficient open-weight mixture-of-experts model (119B total, about 6B active) that accepts text and images. It is Mistral's low-cost option and the replacement for older Small, Magistral Small and Devstral Small models.
Current Open weights
ReleasedMar 16, 2026
Context window256K tokens
Input → outputtext, image → text
API price$0.15 per million input tokens and $0.60 per million output tokens
Consumer access—
Best for
Low-cost, Apache 2.0 licensed model for high-volume chat, reasoning and coding tasks or self-hosting.
Overview
Mistral Small 4 is an efficient open-weight mixture-of-experts model (119B total, about 6B active) that accepts text and images. It is Mistral's low-cost option and the replacement for older Small, Magistral Small and Devstral Small models.
Strengths and limits
- Apache 2.0 license
- 119B MoE with only about 6B active parameters per token, so it is cheap to run
- Configurable reasoning effort; merges Magistral, Pixtral and Devstral capabilities
- Few exact benchmark numbers published by Mistral
- Smaller and less capable than Medium 3.5 and Large 4 on hard tasks
Published benchmark scores
AA LCR0.72
Source: mistral.ai
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →