| DeepSeek V4 | Claude Opus 4.7 | |
|---|---|---|
| arenaElo | 1395 | 1420 |
| humanEval | 91 | 93.1 |
| sweBench | 62 | 64.3 |
| mmlu | 91.5 | 92.3 |
| gpqa | 73 | 65.2 |
| aime | 90 | 88 |
| Price in/out ($/M) | $1.74 / $3.48 | $5 / $25 |
DeepSeek V4 (released April 2026) matches the 1M-token context of Opus 4.7 at roughly 1/6 the price ($1.74/$3.48 vs $15/$75 per M tokens). Opus 4.7 still leads on Chatbot Arena, agentic tool use and coding polish; V4 wins on cost-per-intelligence and is fully self-hostable under MIT license. Pick Opus 4.7 for mission-critical agents, V4 for scale-sensitive STEM and document workloads.