Compare / Devstral Small 2 vs Mistral Small 3.2
Devstral Small 2 vs Mistral Small 3.2
Updated November 2026
Verdict
Devstral Small 2 scores higher on the Small LM Index: 37.3 vs 35.7, although their likely ranges overlap, so the lead is not certain. On the 9 benchmarks both labs report, Devstral Small 2 wins 5 and Mistral Small 3.2 wins 4. Both need about 16.9 GB of memory at 4-bit quantization. Only Mistral Small 3.2 has a hosted API on OpenRouter ($0.11 per million tokens); the other has to be self-hosted.
Pick Devstral Small 2 if…
- you want the strongest model overall
- you mostly write code
- you need math and reasoning
- you work with long documents
Pick Mistral Small 3.2 if…
- you need precise instruction following
- you build agents that call tools
- you want a hosted API rather than running it yourself
Specifications
| Devstral Small 2 | Mistral Small 3.2 | |
|---|---|---|
| Small LM Index | 37.3 (likely 34–40) | 35.7 (likely 33–39) |
| Lab | Mistral AI · Europe | Mistral AI · Europe |
| Released | 2025-12-09 | 2025-06-20 |
| Parameters | 24B (dense) | 24B (dense) |
| Memory at 4-bit | ~16.9 GB | ~16.9 GB |
| Context window | 256K tokens | 128K tokens |
| Input | text, image | text, image |
| Reasoning | No thinking | No thinking |
| License | Apache-2.0 | Apache-2.0 |
| API price / 1M tokens | Self-host only | $0.11 ($0.075 in, $0.20 out) |
| Hugging Face downloads (30 days) | 123.7K | 115.6K |
Benchmarks
Higher is better. Bold marks the winner where both labs report a score. ≈ marks the index model's estimate for a score that wasn't reported, ✓ an independent measurement and * a score measured by another lab.
| Benchmark | Devstral Small 2 | Mistral Small 3.2 |
|---|---|---|
| MMLU-Pro Knowledge | 67.8✓ | 68.1✓ |
| GPQA Diamond Knowledge | 53.2✓ | 50.5✓ |
| Humanity's Last Exam Reasoning | 3.5✓ | 4.3✓ |
| AIME Math | 34.3✓ | 32.3✓ |
| LiveCodeBench Coding | 34.8✓ | 27.5✓ |
| SWE-bench Verified Coding | 68.0 | ≈ 40 |
| Terminal-Bench Agentic | 22.5* | — |
| τ²-Bench Agentic | 23.4✓ | 29.5✓ |
| IFBench Instruction following | 31.2✓ | 33.5✓ |
| MATH-500 Math | — | 88.3✓ |
| AA-LCR Long context | 28.0✓ | 20.3✓ |
| SciCode Coding | 32.4✓ | 28.6✓ |
Scores are mostly reported by each lab with its own settings, so small gaps are noise. See the methodology. Open in the interactive comparison to add more models.
Questions
Which is better, Devstral Small 2 or Mistral Small 3.2?
Devstral Small 2 scores higher on the Small LM Index: 37.3 vs 35.7, although their likely ranges overlap, so the lead is not certain. On the 9 benchmarks both labs report, Devstral Small 2 wins 5 and Mistral Small 3.2 wins 4.
Which is cheaper to run?
Only Mistral Small 3.2 has a hosted API on OpenRouter ($0.11 per million tokens); the other has to be self-hosted. Both need about 16.9 GB of memory at 4-bit quantization.
Which has the longer context window?
Devstral Small 2 has the longer context window: 256K tokens against 128K.
Can I use Devstral Small 2 and Mistral Small 3.2 commercially?
Both are released under the Apache-2.0 license. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.
Related comparisons
- Devstral Small 2 vs Gemma 4 26B-A4B
- Devstral Small 2 vs LFM2.5 24B-A2B
- Devstral Small 2 vs Magistral Small 1.2
- Devstral Small 2 vs EuroLLM 22B Instruct
- LFM2.5 24B-A2B vs Mistral Small 3.2
- Magistral Small 1.2 vs Mistral Small 3.2
- Gemma 4 26B-A4B vs Mistral Small 3.2
- EuroLLM 22B Instruct vs Mistral Small 3.2