Compare / Gemma 4 26B-A4B vs Devstral Small 2
Gemma 4 26B-A4B vs Devstral Small 2
Updated November 2026
Verdict
Gemma 4 26B-A4B scores higher on the Small LM Index: 64.8 vs 37.3. On the 11 benchmarks both labs report, Gemma 4 26B-A4B wins 10 and Devstral Small 2 wins 1. Both need about 16.9 GB of memory at 4-bit quantization. Only Gemma 4 26B-A4B has a hosted API on OpenRouter ($0.068 per million tokens); the other has to be self-hosted.
Pick Gemma 4 26B-A4B if…
- you want the strongest model overall
- you mostly write code
- you need math and reasoning
- you need precise instruction following
- you build agents that call tools
- you want a hosted API rather than running it yourself
- you want the more widely used model (more quantized builds and community support)
Pick Devstral Small 2 if…
No clear advantage in the data we have.
Specifications
| Gemma 4 26B-A4B | Devstral Small 2 | |
|---|---|---|
| Small LM Index | 64.8 (likely 63–67) | 37.3 (likely 34–40) |
| Lab | Google · United States | Mistral AI · Europe |
| Released | 2026-04-02 | 2025-12-09 |
| Parameters | 25.2B (3.8B active, MoE) | 24B (dense) |
| Memory at 4-bit | ~16.8 GB | ~16.9 GB |
| Context window | 256K tokens | 256K tokens |
| Input | text, image | text, image |
| Reasoning | Hybrid (thinking on/off) | No thinking |
| License | Apache-2.0 | Apache-2.0 |
| API price / 1M tokens | $0.068 ($0.040 in, $0.15 out) | Self-host only |
| Hugging Face downloads (30 days) | 5.8M | 123.7K |
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 | Gemma 4 26B-A4B | Devstral Small 2 |
|---|---|---|
| MMLU-Pro Knowledge | 82.6 | 67.8✓ |
| GPQA Diamond Knowledge | 79.2✓ | 53.2✓ |
| Humanity's Last Exam Reasoning | 19.3✓ | 3.5✓ |
| AIME Math | 88.3 | 34.3✓ |
| LiveCodeBench Coding | 77.1 | 34.8✓ |
| SWE-bench Verified Coding | 17.4* | 68.0 |
| Terminal-Bench Agentic | 34.2* | 22.5* |
| τ²-Bench Agentic | 43.6✓ | 23.4✓ |
| IFBench Instruction following | 72.4✓ | 31.2✓ |
| MMMU-Pro Vision | 73.8 | — |
| AA-LCR Long context | 65.7✓ | 28.0✓ |
| SciCode Coding | 40.0✓ | 32.4✓ |
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, Gemma 4 26B-A4B or Devstral Small 2?
Gemma 4 26B-A4B scores higher on the Small LM Index: 64.8 vs 37.3. On the 11 benchmarks both labs report, Gemma 4 26B-A4B wins 10 and Devstral Small 2 wins 1.
Which is cheaper to run?
Only Gemma 4 26B-A4B has a hosted API on OpenRouter ($0.068 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?
Both have a 256K-token context window.
Can I use Gemma 4 26B-A4B and Devstral Small 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.