ClawForge: leaderboard
Interactive agent benchmarks face a tension between scalable construction and realistic workflow evaluation. Hand-authored tasks are expensive to extend and revise, while static prompt evaluation misses failures that only appear when agents operate over persistent state. Existing interactive benchmarks have advanced agent evaluation significantly, but most initialize tasks from clean state and do
Metric: Strict Acc. (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 45.3 | #63 |
| 2 | Claude Sonnet 4.6 | 42.3 | #90 |
| 3 | GPT-5.4 | 40.9 | #79 |
| 4 | GPT-5.3 Codex | 40.6 | #69 |
| 5 | O3 | 39.5 | #124 |
| 6 | GPT-5.2 | 38.7 | #110 |
| 7 | Kimi K2.5 | 35.9 | #144 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clawforge · How It Works · Data refreshed daily, snapshot 2026-10-20.