Anthropic
Claude Opus 4
Start of the Claude 4 generation, released with Sonnet 4. Aimed at coding, reasoning and agents.
- Announced
- Access
- Public
Benchmark scores
| Test | Score | Best on the tracker | How it was measured |
|---|---|---|---|
| Epoch Capabilities Indexmany tests combined into one score | 143 points | GPT-6.2 Astra 174 points | Independentlisted as "Claude Opus 4", CI 141.26 to 144.48Run 9 Nov 2026 |
| Humanity's Last Examexpert questions, no tools | 11% | GPT-6 Astra 55% | Independentleaderboard entry "Claude Opus 4 (Thinking)" (claude-opus-4-20250514_unknown) |
| GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated | 76% | Gemini 4 Argon 96% | Independent16K thinking budget (claude-opus-4-20250514_16K)Run 22 May 2025 |
| ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated | 8.6% | GPT-6 Astra 95% | IndependentARC Prize config "Claude Opus 4 (Thinking 16K)" (claude-opus-4-20250514-thinking-16k) |
| OTIS Mock AIMEcompetition mathsSaturated | 64% | GPT-5.5 100% | Independent27K thinking budget (claude-opus-4-20250514_27K)Run 28 May 2025 |
| SWE-bench Verifiedfixing real software bugsKnown flaws | 71% | Claude Opus 4.7 83% | Independentdefault settings (claude-opus-4-20250514)Run 6 Feb 2026 |
| METR time horizontask length an AI agent finishes half the timeNo longer updated | 1.7 h | Claude Mythos Preview 17 h | IndependentMETR Horizon v1.1 run 'claude_4_opus_inspect'; 95% CI 60.0–163.5 min |
Sources
- Introducing Claude 4Anthropic · Official · anthropic.com
- Epoch AI Benchmarking Hub (model release dates and runs)Epoch AI · Dataset · 9 Nov 2026 · epoch.ai