SQBench: leaderboard
SQBench (Shaqiu Community, 2026) evaluates language-model agents on 220 production-oriented delivery tasks that require processing input assets, using tools and producing a specified deliverable. A task passes only when the deliverable is functionally complete and triggers no penalty in a 10-dimension risk matrix; the primary score weights the pass rates of the three task layers 0.2, 0.6 and 0.2, in percent, higher is better.
Metric: Weighted Strict Pass@1 (%; 0.2 times L1 plus 0.6 times L2 plus 0.2 times L3 strict pass rates, the prespecified primary aggregate 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 | 60.5 | #54 |
| 2 | Claude Opus 4.8 (Max) | 57.6 | #56 (Claude Opus 4.8) |
| 3 | GLM-5.2 (Max) | 52.7 | #138 (GLM-5.2) |
| 4 | Qwen 3.7 Max | 51.1 | #145 |
| 5 | Kimi K2.7 Code | 50.4 | #185 |
| 6 | DeepSeek V4 Pro (Max) | 49.5 | #188 (DeepSeek V4 Pro) |
| 7 | MiMo-V2.5 | 48.5 | #354 |
| 8 | DeepSeek V4 Pro (High) | 48.3 | #188 (DeepSeek V4 Pro) |
| 9 | Doubao-Seed-2.1-Pro (High) | 48.1 | |
| 10 | Kimi K2.6 | 47.3 | #178 |
| 11 | DeepSeek V4 Flash (High) | 47.1 | #258 (DeepSeek V4 Flash) |
| 12 | GLM-5.1 | 46.7 | #179 |
| 13 | MiniMax-M3 | 46.7 | #269 |
| 14 | Qwen 3.6 Plus | 45.4 | #222 |
| 15 | Qwen 3.7 Plus | 45.3 | #207 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sqbench · How It Works · Data refreshed daily, snapshot 2026-11-13.