SaaSBench (Engineering) - Logic Capability: leaderboard
The Logic capability dimension of SaaSBench (Ren et al., 2026), pooled over all 30 SaaS systems a coding agent builds from scratch: whether domain business behaviour such as orders, billing, moderation and workflows is implemented. Share of the validation-node score earned in that dimension, in percent; higher is better.
Metric: Capability score (%; share of the Logic (domain business behaviour) validation-node max score earned across all 30 tasks, pass@1; agent builds and deploys an enterprise SaaS system from a PRD). Source: arxiv.org. Saturation forecast: Around 2029. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 8.79 | #75 |
| 2 | GLM-5.1 | 7.81 | #175 |
| 3 | Gemini 3.1 Pro (Preview) | 6.77 | #91 |
| 4 | DeepSeek V4 Pro | 6.13 | #181 |
| 5 | GPT-5.4 (xHigh) | 5.97 | #121 (GPT-5.4) |
| 6 | Kimi K2.6 | 5.85 | #173 |
| 7 | MiniMax-M2.7 | 3.8 | #519 |
| 8 | Qwen 3.6 Plus | 1.02 | #217 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=saasbench-engineering-logic-capability · How It Works · Data refreshed daily, snapshot 2026-11-12.