Claude Opus 4.8 (Thinking): benchmark results
Claude Opus 4.8 evaluated with thinking enabled. Provider: Anthropic. Released 2026-05-28. Access: API.
Aggregate Rating 1773 ± 16, rank #193 of 2170 rated models, from 61 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ComboShoppingBench - Response Quality | 96.6 | Pass rate (%; LLM-judged) | 100 |
| WorkBuddy Bench (Claude Code) - Code | 77.9 | Score (0-100) on the 80 repository-level software engineerin | 100 |
| WorkBuddy Bench (Claude Code) - Web | 69.86 | Score (0-100) on the 70 front-end and GUI tasks scored by ru | 100 |
| WorldCup Arena | 897 | Total points (out of 1790); a seven-market prediction card f | 100 |
| WorldCup Arena - Match Outcome | 66.3 | Accuracy (%; home, draw or away pick on the 90-minute result | 100 |
| Kagi LLM Benchmark | 88.8 | Accuracy (%) | 98.6 |
| ReLE - Reasoning and Mathematics | 89.9 | Accuracy (%) | 98.6 |
| ComboShoppingBench - Coupon-ID Validity | 100 | Pass rate (%) | 97.6 |
| Game2Real-Bench - Toy Game - Constraint Reasoning | 62 | Accuracy (%) | 95.8 |
| ReLE - Education - Primary School Subjects | 70.7 | Accuracy (%) | 94.9 |
| ComboShoppingBench - Claim Faithfulness | 94.5 | Pass rate (%; LLM-judged) | 92.9 |
| ComboShoppingBench - Coupon Legality | 98.6 | Pass rate (%) | 92.9 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-8-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.