UniClawBench - Long Context: leaderboard
Metric: Pass rate (%) on the 80 long-context tasks of 400 bilingual real-world tasks run live in Docker containers under the OpenClaw framework (v2026.3.11), with up to two simulated user follow-ups and a hidden GPT-5.4 Codex supervisor that grades step checkpoints. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 | 43.8 |
| 2 | Claude Sonnet 4.6 | 40 |
| 3 | Kimi K2.6 | 28.7 |
| 4 | GPT-5.4 | 22.5 |
| 5 | Qwen 3.5 Plus | 20 |
| 6 | Gemini 3.1 Pro (Preview) | 18.8 |
| 7 | Gemini 3 Flash | 16.3 |
| 8 | GPT-5.4 Mini | 10 |
| 9 | GPT-4.1 | 7.5 |
| 10 | Gemini 3.1 Flash Lite | 6.2 |
Interactive version: theaggregate.ai/benchmark?slug=uniclawbench-long-context · How It Works · Data refreshed daily, snapshot 2026-09-29.