SCHEDBench (Full-Variation): leaderboard
SCHEDBench (2026) verbalizes canonical job-shop, project-scheduling, nurse-rostering and timetabling instances as natural-language problems and checks whether large language models produce constraint-feasible schedules. Feasibility is the share of instances whose output satisfies all hard constraints under an exact OR-Tools verifier, in percent, higher is better, on the full surface-form rendering with themed names, paraphrased templates and reordered constraints.
Metric: Feasibility rate (%; share of SCHEDBench natural-language scheduling instances whose model output satisfies all hard constraints, verified by OR-Tools CP-SAT; six canonical job-shop, project-scheduling, nurse-rostering and timetabling families; greedy decoding, zero-shot, no external solver; all three surface-form axes (themed names, paraphrased templates, constraint reordering) active). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (2026-04-23) | 55.9 | #33 |
| 2 | Gemini 3 Flash (Preview) | 29.5 | #141 |
| 3 | Qwen 3.5 27B | 25 | #358 |
| 4 | Qwen 3.5 397B A17B | 24.3 | #242 |
| 5 | Qwen 3.5 122B A10B | 14.3 | #329 |
| 6 | Claude Sonnet 4.6 | 12.8 | #158 |
| 7 | Claude Opus 4.6 | 8.9 | #100 |
| 8 | Claude Haiku 4.5 | 4.9 | #568 |
| 9 | Llama 3.3 70B | 4.9 | #1511 |
| 10 | Llama 4 Maverick | 4.5 | #1105 |
| 11 | GPT-5.4 Mini (2026-03-17) | 4.3 | #285 |
| 12 | Gemini 3.1 Flash Lite | 3.9 | #363 |
| 13 | GPT-5.4 (2026-03-05) | 3.7 | #84 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=schedbench-full-variation · How It Works · Data refreshed daily, snapshot 2026-11-12.