Text2Opt-Bench: leaderboard
Natural-language optimization modeling from Text2Opt-Bench (Gao, Ge, Berenbeim, Bastian and Sala, 2026): the model reads the problem and its instance data in natural language and writes Gurobi code, and an answer counts when the solved objective matches the verified optimum. Pass@1 accuracy in percent averaged over eleven template categories of 50 problems each (transportation, disaster response, job-shop scheduling, vehicle routing, project scheduling, facility location, power transmission, queuing, stochastic, multi-objective and modified facility location); higher is better.
Metric: Pass@1 accuracy (%; mean over the eleven template categories, 50 small-tier instances each, 550 problems; solver-verified optimal objective). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 87.6 | #158 |
| 2 | Claude Opus 4.6 | 86.7 | #100 |
| 3 | GPT-5 | 86.2 | #163 |
| 4 | O4 Mini | 80.4 | #354 |
| 5 | DeepSeek R1 | 72.9 | #467 |
| 6 | DeepSeek V3.2 | 72 | #403 |
| 7 | GPT-5 Nano | 59.1 | #888 |
| 8 | Llama 3.3 70B Instruct | 35.1 | #1353 |
| 9 | Qwen 2.5 7B Instruct | 4.5 | #2583 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=text2opt-bench · How It Works · Data refreshed daily, snapshot 2026-11-12.