MODELS
Mistral Small 3.1
Open-weights 24B with vision and a 128K window at bargain prices.
Specs
- Context window
- 128,000
- Max output
- 8,192
- Modalities
- text, image
- Tool use
- ✓
- Vision
- ✓
- Streaming
- ✓
- License
- apache-2.0
- Released
- 2025-03-17
Pricing
- Input / 1M
- $0.10
- Output / 1M
- $0.30
Cost estimate
Mistral Small 3.1 is a 24B Apache-2.0 model that handles text and images, supports tool calls, and runs a 128K context. At $0.10/$0.30 per million tokens it sits in the same price tier as Gemini Flash and GPT-4o-mini, but you can also self-host the weights. Aimed at builders who want a capable multimodal workhorse without per-call lock-in.
Editor's verdict
Pick this when you need a small multimodal model you can actually run on your own GPUs—it's one of the few open-weight options at this size with real vision support. Raw quality trails GPT-4o-mini and Gemini 2.0 Flash on hard reasoning and OCR-heavy tasks, and it's not a thinking model, so don't expect deep chains of thought. The sweet spot is high-volume document or image classification where ownership matters more than the last few benchmark points.
Reviews
No reviews yet. Be the first.
Last updated: 2026-04-29