SQBench - L2 Composite Skills: leaderboard
The composite-skill layer of SQBench (Shaqiu Community, 2026): the percent of its 60 L2 delivery tasks that an agent completes with no triggered risk penalty, one run per task. Higher is better.
Metric: Strict Pass@1 (%; the 60 L2 composite-skill tasks on SQBench v1.0, 220 production-oriented delivery tasks in which an agent processes input assets, uses tools and produces a specified deliverable; a task passes only with functional Completion 1 and no triggered 10-dimension risk penalty; one run per configuration and task; higher is better). Source: arxiv.org. Saturation forecast: Around February 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 | 71.7 | #54 |
| 2 | Claude Opus 4.8 (Max) | 66.7 | #56 (Claude Opus 4.8) |
| 3 | Kimi K2.7 Code | 63.3 | #185 |
| 4 | Qwen 3.7 Max | 61.7 | #145 |
| 5 | GLM-5.2 (Max) | 61.7 | #138 (GLM-5.2) |
| 6 | Doubao-Seed-2.1-Pro (High) | 61.7 | |
| 7 | MiniMax-M3 | 60 | #269 |
| 8 | MiMo-V2.5 | 60 | #354 |
| 9 | DeepSeek V4 Pro (Max) | 60 | #188 (DeepSeek V4 Pro) |
| 10 | DeepSeek V4 Pro (High) | 60 | #188 (DeepSeek V4 Pro) |
| 11 | DeepSeek V4 Flash (High) | 60 | #258 (DeepSeek V4 Flash) |
| 12 | Kimi K2.6 | 58.3 | #178 |
| 13 | Qwen 3.6 Plus | 56.7 | #222 |
| 14 | GLM-5.1 | 56.7 | #179 |
| 15 | MiMo-V2.5-Pro | 56.7 | #279 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sqbench-l2-composite-skills · How It Works · Data refreshed daily, snapshot 2026-11-13.