Claude Opus 5.5
Still first on Terminal-Bench, WebDev, and the Android long board. On sudo L7 it ties Sonnet 5.5 at 45%, and the page lists Opus first. Cheaper models fit batch jobs.
Anthropic · Proprietary · Context 1M · $4 / $20 per million tokens · Usual agent Claude Code · Fetched 2026-10-10
Benchmark scores
Boards older than 90 days are not counted as recommendationsBenchmarkSettingScoreRankCostAs of
Use it or not
Why
- First on Terminal-Bench, LMArena WebDev, and the Android Bench long-horizon board.
- On Terminal-Bench it leads Sonnet 5.5 by three points (64.8% vs 61.8%) and the run costs less ($4.7k vs $7.3k).
When not
- For batch or CI work, GPT-6.1 Sol ties for 3rd on Terminal-Bench at 58.2% and $634.24.
- For frontend only, Claude Sonnet 5.5 scores 1774 on WebDev at half the token price ($2 / $10).
- Latency: the Android Bench long run averages 15.2 hours.
- Production judgment: tied with Sonnet on sudo L7. The blog puts a task at about $10, and Sonnet at about $14.