Claude Opus 4.8 (xHigh): benchmark results
Claude Opus 4.8 evaluated at the xhigh reasoning-effort setting. Provider: Anthropic. Released 2026-05-28. Access: API.
Aggregate Rating 1858 ± 13, rank #60 of 2168 rated models, from 116 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Business Utility Eval (Claude Code) | 0.42 | Business Utility (0-1) | 100 |
| GDPevo | 50.63 | Mean accuracy (%), base, 24 task groups, three repetitions | 100 |
| GDPevo - fewshot | 67.07 | Mean accuracy (%), fewshot, 24 task groups, three repetition | 100 |
| GDPevo - reflect-3 | 59.27 | Mean accuracy (%), reflect-3, 24 task groups, three repetiti | 100 |
| Nejumi 4 - jaster (0-shot) - JaMP | 85 | Exact match (%) | 100 |
| Nejumi 4 - jaster (2-shot) - ALT English-to-Japanese | 92.75 | COMET wmt22 (x100) | 100 |
| PTXBench (B200) - MHA Forward | 38.5 | Multi-head attention forward pass on B200: target-instructio | 100 |
| PTXBench (B200) - MHA Forward Causal | 32.3 | Causal multi-head attention forward pass on B200: target-ins | 100 |
| PTXBench (H100) - GEMM | 94.8 | GEMM on H100: target-instruction turn correctness rate (%) w | 100 |
| PTXBench (H100) - MHA Backward | 79.2 | Multi-head attention backward pass on H100: target-instructi | 100 |
| PTXBench (H100) - MHA Forward | 90.6 | Multi-head attention forward pass on H100: target-instructio | 100 |
| PTXBench (H100) - MHA Forward Causal | 75 | Causal multi-head attention forward pass on H100: target-ins | 100 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-8-xhigh · How It Works · Data refreshed daily, snapshot 2026-10-20.