Claw-Anything - Consistency (Pass^3): leaderboard

Claw-Anything (2026) benchmarks always-on personal assistant agents in a simulated digital world with months of user activity, interdependent backend services and CLI and GUI devices, all run in the OpenHarness agent scaffold. This board is the share in percent of tasks solved in all three runs; higher is better.

Metric: Pass^3 (%): share of tasks solved in all three runs, over the 200 Claw-Anything tasks (simulated months of user activity, interdependent backend services, CLI Linux and GUI Android devices, reactive and proactive tasks), every model in the OpenHarness agent scaffold, three independent runs, rubric checks plus a Claude Sonnet 4.5 judge giving a pass label and a soft score; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.520#54
2GLM-5.117#198
3Claude Opus 4.713.5#85
4Claude Sonnet 4.512#261
5Kimi K2.66.5#192
6Qwen 3.6 27B6#351
7MiniMax-M2.73.5#539
8Qwen 3.5 27B2#375

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=claw-anything-consistency-pass-pow-3 · How It Works · Data refreshed daily, snapshot 2026-11-21.