ClawMark: leaderboard
Metric: Weighted score (0-100): mean over tasks of the weight-normalized checker pass rate; 100 multi-day coworker tasks in 13 professional scenarios on five stateful sandboxed services (filesystem, email, calendar, knowledge base, spreadsheet) with multimodal evidence; every model runs in the OpenClaw agent framework, one sweep, deterministic Python checkers (no LLM judge); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 (Thinking) | 75.8 |
| 2 | Claude Opus 4.6 (Thinking) | 74.6 |
| 3 | GPT-5.4 (High) | 72 |
| 4 | Kimi K2.6 | 68.4 |
| 5 | Gemini 3.1 Pro (Preview) | 68.2 |
| 6 | Qwen 3.6 Plus | 57.2 |
| 7 | Kimi K2.5 | 56 |
Interactive version: theaggregate.ai/benchmark?slug=clawmark · How It Works · Data refreshed daily, snapshot 2026-10-07.