Claude Mythos Preview: benchmark results
Anthropic preview Claude model for high-end reasoning, coding, and general tasks. Provider: Anthropic. Released 2026-04-07. Access: Private.
Aggregate Rating 1984 ± 12, rank #4 of 1656 rated models, from 97 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Anthropic Claude Code Malicious Requests - Refusal Rate (Sonnet 5 & Opus 5 System Cards) | 96.21 | Refusal Rate (%) | 100 |
| Anthropic ECI (AECI) | 159.2 | AECI Score | 100 |
| BBQ Ambiguous Accuracy (Sonnet 5 & Opus 5 System Cards) | 100 | Accuracy (%) | 100 |
| BenchGecko Score | 99.7 | Composite Score | 100 |
| BioMysteryBench Verified - Human Solvable (Anthropic) | 82.6 | Score (%) | 100 |
| BioPipelineBench Verified | 88.1 | Accuracy (system card) | 100 |
| BioPipelineBench Verified (Anthropic) | 88.1 | Score (%) | 100 |
| ExploitBench v8-bench | 72 | Capability coverage (%), all-runs view with reported seed/tu | 100 |
| ExploitBench v8-bench (AutoNudge) | 78 | Capability coverage (%), all-runs view with reported seed/tu | 100 |
| LAB-Bench FigQA (Mythos Preview No Tools) | 79.7 | Accuracy (%) | 100 |
| LAB-Bench FigQA (Mythos Preview Tools) | 89 | Accuracy (%) | 100 |
| LABBench2 - Clinical Trial Questions (Anthropic) | 86.3 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-mythos-preview · How It Works · Data refreshed daily, snapshot 2026-10-20.