Mistral Small 4
Open SourceMistral AI•Released on 2026-03-16
Mistral's unified model combining instruct, reasoning (Magistral), coding (Devstral), and multimodal (Pixtral) capabilities. 119B total / 6B active MoE parameters. Apache 2.0 license. 256K context. Configurable reasoning_effort parameter for balancing speed vs depth.
80
Overall Score
Voice of the community
sample 245“Mistral Small 4 with Apache 2.0 is huge. One model for reasoning, vision, and coding - exactly what I needed.”
“The reasoning_effort parameter is brilliant. Fast mode for simple tasks, high for complex problems.”
Core Specs
262K
Context Window
33K
Max Output
ReasoningOpen Sourcetextimage
Scenario Scores
Pros & Cons
Sentiment72% +20% ·8% −
Pros
- +Apache 2.0 open source
- +Excellent price ($0.15/$0.60)
- +Unified reasoning + multimodal + coding
- +Configurable reasoning effort
- +40% faster than Mistral Small 3
Cons
- −Requires high-end GPUs (4x H100 minimum)
- −Behind frontier models on hardest benchmarks
- −New model, less community validation
Pricing
Input (per 1M tokens)$0.15
Output (per 1M tokens)$0.60
Free trial available
Updated on 2026-04-01
Get Started
1Visit the provider's website
2Create an account
3Start using the model
Benchmarks
mmlu85%
humanEval88%
liveCodeBench72%
Reliability
Incidents (30d)0
开源模型，可靠性取决于托管方