Agentic Task Execution on CAPTRACEBENCH Easy 1.0
81.4Success Rate (SR)Claude-Opus-4.6
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude-Opus-4.6Setting=w/ Oracle Skills2026.06 | 81.4 | 91.1 | |
| Claude-Opus-4.6Setting=Raw Traces (w/ Extracted Skills)2026.06 | 80 | 87 | |
| Claude-Opus-4.6Setting=Raw Traces (w/ Evolved Skills)2026.06 | 74.5 | 86 | |
| Claude-Opus-4.6Setting=Raw Traces (w/ Retrieval Reuse)2026.06 | 73.8 | 86 | |
| Claude-Opus-4.6Setting=REDACT Protected (w/ Extracted Skills)2026.06 | 72.5 | 80.9 | |
| Gemini-3-ProSetting=No Skills2026.06 | 72.3 | 83.6 | |
| GPT-5.2-CodexSetting=No Skills2026.06 | 71.3 | 86.7 | |
| Claude-Sonnet-4.6Setting=No Skills2026.06 | 70.5 | 80.6 | |
| Average.Setting=No Skills2026.06 | 69.6 | 82.9 | |
| Average.Setting=REDACT Protected (w/ Extracted Skills)2026.06 | 69.3 | 80 | |
| Gemini-3-FlashSetting=No Skills2026.06 | 69.2 | 83.4 | |
| Claude-Opus-4.6Setting=No Skills2026.06 | 68.8 | 84 | |
| Claude-Opus-4.6Setting=REDACT Protected (w/ Evolved Skills)2026.06 | 67.6 | 82.1 | |
| Claude-Opus-4.6Setting=REDACT Protected (w/ Retrieval Reuse)2026.06 | 66.8 | 81.7 | |
| Claude-Haiku-4.5Setting=No Skills2026.06 | 65.3 | 79.1 |