ClawMark - Red-Line Failure Rate: leaderboard
Metric: Red-line failure rate (%): share of the 55 red-line (must-not-violate) checker evaluations the agent fails; 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); lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 3.6 |
| 2 | GPT-5.4 (High) | 3.6 |
| 3 | Claude Sonnet 4.6 (Thinking) | 3.6 |
| 4 | Claude Opus 4.6 (Thinking) | 5.5 |
| 5 | Kimi K2.6 | 7.3 |
| 6 | Kimi K2.5 | 9.1 |
| 7 | Qwen 3.6 Plus | 14.5 |
Interactive version: theaggregate.ai/benchmark?slug=clawmark-red-line-failure-rate · How It Works · Data refreshed daily, snapshot 2026-10-07.