Claude Sonnet 5
Anthropic·
LLMsmidcloud
Overview
Sonnet-tier Claude model for agentic coding, computer use and tool use, released as the new default in the Claude apps and the API.
Capabilities and innovations
Agentic codingComputer useExtended thinkingTool useVision inputAPI: claude-sonnet-5OSWorld-Verified computer useIntroductory pricing through 2026-08-31
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 88.89% | independent GPQA Diamond 88,89%, measured by Vals AI. Artificial Analysis measures 91,11% at max effort. |
| Humanity’s Last Exam (no tools) | 43.2% | vendor Humanity's Last Exam ohne Tools (43,2%); mit Tools 57,4%. |
| SWE-bench Pro | 63.2% | vendor |
| MMLU-Pro | 87.55% | independent MMLU Pro 87,55%, measured by Vals AI. |
| Terminal-Bench 2.x | 74.53% | independent Terminal-Bench 2.1 74.53%, measured by Vals AI with the Terminus 2 harness (accessed 2026-09-23). Vals AI is the reference harness for this field so values compare across vendors; vendor and other third-party figures stay in the notes below. Replaces 80.4 (vendor). |
| Terminal-Bench 3.0 | 14.6% | independent Terminal-Bench 3.0 14.6% (± 2.9, 95% CI), official leaderboard maintained by the Terminal-Bench team (Stanford / Harbor / Laude Institute), max effort, agent harness: Claude Code (accessed 2026-09-23). Do not compare against the terminal_bench column: 3.0 is a harder generation on a different scale. |
| Terminal-Bench 4.0 | 12.4% | independent Terminal-Bench 4.0 12.4% (± 3.1, 95% CI), official leaderboard maintained by the Terminal-Bench team (Stanford / Harbor / Laude Institute), max effort, agent harness: Claude Code (accessed 2026-09-23). 4.0 is its own generation and not comparable with the 2.x or 3.0 fields. |
| MMMU-Pro | 77.3% | third party MMMU-Pro 77,3%, measured by Artificial Analysis (Adaptive Reasoning, max effort). |
Links
More from Anthropic
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.