Kimi K3 Tech Report - 24/7 ClawBench 2.0: leaderboard
Moonshot AI's in-house 24/7 ClawBench 2.0 simulates always-on assistant work in which tasks span multiple days, events arrive concurrently and interruptions are routine; all models run in OpenClaw. Kimi K3 technical report, Table 3; score in percent, higher is better. Claude Fable 5's score includes 2 tasks it refused.
Metric: Score (%). Source: raw.githubusercontent.com. Saturation forecast: Around 2029. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 52 | #22 (GPT-5.6 Sol) |
| 2 | GPT-5.5 (xHigh) | 48.5 | #33 (GPT-5.5) |
| 3 | Kimi K3 (Max) | 48.3 | #35 (Kimi K3) |
| 4 | Claude Fable 5 (Max) | 47.4 | #20 (Claude Fable 5) |
| 5 | Claude Opus 4.8 (Max) | 47.2 | #39 (Claude Opus 4.8) |
| 6 | GLM-5.2 (Max) | 43.2 | #80 (GLM-5.2) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kimi-k3-tech-report-24-7-clawbench-2-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.