Claude Opus 4.1 (Thinking): benchmark results
Claude Opus 4.1 evaluated with thinking enabled. Provider: Anthropic. Released 2025-08-05. Access: API.
Aggregate Rating 1690 ± 15, rank #390 of 2168 rated models, from 110 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - JTruthfulQA | 78.01 | Score (%) | 100 |
| SuperCLUE General (September 2025) - Hallucination Control | 85.23 | Score | 100 |
| Nejumi 4 - Toxicity - Violation Categories | 61.11 | Criteria met (%) | 99.3 |
| Nejumi 4 - BFCL - Multi-Turn | 66.67 | Accuracy (%) | 99 |
| BenchTable | 82.7 | Total Score (%) | 98.8 |
| BenchTable - Reasoning | 90.4 | Weighted Score (%) | 98.8 |
| BenchTable - STEM | 89.6 | Weighted Score (%) | 98.8 |
| BenchTable - Tech | 93.9 | Weighted Score (%) | 98.8 |
| AA MMLU-Pro | 87.99 | Accuracy (%) | 98.3 |
| Nejumi 4 - MT-Bench (Japanese) - Extraction | 99 | Judge rating (1-10, x10) | 98.3 |
| Nejumi 4 - jaster (0-shot) - ALT English-to-Japanese | 92.58 | COMET wmt22 (x100) | 98.3 |
| Nejumi 4 - jaster (2-shot) - JBLiMP | 97 | Exact match (%) | 98.3 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-1-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.