| DeepSeek V3 | Llama 4 Maverick | |
|---|---|---|
| arenaElo | 1355 | 1322 |
| humanEval | 86 | 82 |
| sweBench | 42 | 45 |
| mmlu | 88.5 | 85.1 |
| gpqa | 59.1 | 54.3 |
| aime | 59.4 | 60 |
| Price in/out ($/M) | $0.26 / $1.03 | $0.2 / $0.8 |
DeepSeek V3 is cheaper and beats Llama 4 on most chat benchmarks; Llama 4 Maverick has a 1M context window and stronger vision. For high-volume text chat, DeepSeek wins on cost. For vision-heavy or long-document workloads, Llama wins.