Agentic Task Execution on CAPTRACEBENCH Medium 1.0
61.5Success RateClaude-Opus-4.6
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Claude-Opus-4.6Setting=w/ Oracle Skills2026.06 | 61.5 | 81.5 | |
| Average.Setting=Raw Traces (w/ Extracted Skills)2026.06 | 48.6 | 74.2 | |
| Claude-Opus-4.6Setting=REDACT Protected (w/ Extracted Skills)2026.06 | 43.2 | 67.2 | |
| Claude-Opus-4.6Setting=No Skills2026.06 | 42.3 | 71.9 | |
| Gemini-3-ProSetting=No Skills2026.06 | 41.8 | 67.9 | |
| GPT-5.2-CodexSetting=No Skills2026.06 | 41.3 | 71.5 | |
| Average.Setting=REDACT Protected (w/ Extracted Skills)2026.06 | 39 | 65.9 | |
| Gemini-3-FlashSetting=No Skills2026.06 | 37.8 | 67 | |
| Average.Setting=No Skills2026.06 | 37.8 | 67.5 | |
| Claude-Sonnet-4.6Setting=No Skills2026.06 | 36.2 | 66.1 | |
| Claude-Haiku-4.5Setting=No Skills2026.06 | 27.5 | 60.9 |