| Claude Mythos | Claude Opus 4.8 | |
|---|---|---|
| arenaElo | 1478 | 1435 |
| humanEval | 98.7 | 94.5 |
| sweBench | 93.9 | 67 |
| mmlu | 94.2 | 93 |
| gpqa | 94.6 | 67.5 |
| aime | 97.6 | 90 |
| Price in/out ($/M) | $25 / $125 | $5 / $25 |
Mythos posts higher raw benchmarks but stays a gated research preview at 5× the price. Opus 4.8 is the practical frontier you can actually ship on, generally available across Anthropic API, Bedrock and Vertex AI at $5 / $25 per M tokens, leading agentic coding and computer use. Pick Mythos only if Anthropic green-lit your use case; otherwise Opus 4.8 is the default for production.