SCHEDBench (Full-Variation): leaderboard
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 September 2028. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 29.5 |
| 2 | Qwen 3.5 27B | 25 |
| 3 | Qwen 3.5 397B A17B | 24.3 |
| 4 | Qwen 3.5 122B A10B | 14.3 |
| 5 | Claude Sonnet 4.6 | 12.8 |
| 6 | Claude Opus 4.6 | 8.9 |
| 7 | Claude Haiku 4.5 | 4.9 |
| 8 | Llama 3.3 70B | 4.9 |
| 9 | Llama 4 Maverick | 4.5 |
| 10 | Gemini 3.1 Flash Lite | 3.9 |
| 11 | GPT-5.4 (2026-03-05) | 3.7 |
Interactive version: theaggregate.ai/benchmark?slug=schedbench-full-variation · How It Works · Data refreshed daily, snapshot 2026-09-26.