| Claude Mythos 5 | Claude Opus 4.8 | |
|---|---|---|
| arenaElo | 1493 | 1435 |
| humanEval | 98 | 94.5 |
| sweBench | 78 | 67 |
| mmlu | 95.1 | 93 |
| gpqa | 92 | 67.5 |
| aime | 94.5 | 90 |
| Price in/out ($/M) | $10 / $50 | $5 / $25 |
Mythos 5 is the most capable Claude, but it stays gated behind Project Glasswing at $10 / $50 per M tokens with always-on thinking and 30-day retention. Opus 4.8 is the practical flagship you can ship on today: generally available at $5 / $25, leading agentic coding and computer use. Pick Mythos 5 only if you have a Glasswing seat and your use case needs the absolute frontier; otherwise Opus 4.8 (or the GA Fable 5) is the default.
Claude Mythos 5 · Claude Opus 4.8 · All AI model comparisons