| Llama 4 Scout | Mistral Small 3 | |
|---|---|---|
| arenaElo | 1278 | 1266 |
| humanEval | 76.3 | 84.8 |
| sweBench | 35 | 30 |
| mmlu | 80.4 | 81.2 |
| gpqa | 49.1 | 45.3 |
| aime | 50 | 45 |
| Price in/out ($/M) | $0.18 / $0.6 | $0.2 / $0.6 |
Scout gives you a 10M-token context. Mistral Small can't match that. Small 3 wins on reasoning and has an Apache 2.0 license. For bulk retrieval, Scout. For self-hosted classification and routing under tight hardware, Small 3.