Mistral Small 3.1
24B Apache-2.0 multimodal model that runs on a single GPU
About
Mistral Small 3.1, released March 17, 2025, is a 24B model with text and image input, a 128K context window, and the Apache 2.0 license, which is unusually permissive at this capability level. Mistral reports it beating Gemma 3 and GPT-4o-mini on several benchmarks, with 80.6% on MMLU and strong multimodal scores on ChartQA and AI2D. It runs on a single RTX 4090 or a 32GB Mac, making it one of the most practical self-hosted vision models. Inference is fast at a reported 150 tokens per second. It is not a reasoning model, so heavy multi-step visual math still favors larger systems.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
MMLU 80.6%, HumanEval ~88%, 150 tok/s inference (Mistral-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Release date 2025-03-17, 24B params, 128K context, Apache 2.0 verified across multiple trackers and Mistral's announcement; MMLU 80.6%, 150 tok/s, single-RTX-4090 claim are Mistral/reporter figures; labeled as such; API pricing $0.10/$0.30 per M tokens from a third-party tracker; attributed.