OptiVerse - Mathematical Programming: leaderboard
Metric: Accuracy (%) on the 367 mathematical programming problems of OptiVerse; the model writes and runs solver code (gurobi, pyomo, cvxpy, ortools and others) and a problem counts as solved only when every required variable and the objective match the ground truth within 0.1% relative error, extracted and checked by a DeepSeek-V3.2-Chat judge; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 58.04 |
| 2 | GPT-5.2 (Thinking) | 55.86 |
| 3 | Gemini 3 Flash | 54.77 |
| 4 | Gemini 2.5 Pro | 53.68 |
| 5 | DeepSeek V3.2 (Thinking) | 53.68 |
| 6 | Claude Sonnet 4.5 (Thinking) | 53.41 |
| 7 | O3 | 52.04 |
| 8 | DeepSeek V3.2 (Non-reasoning) | 51.23 |
| 9 | O4 Mini | 50.95 |
| 10 | Gemini 2.5 Flash | 46.05 |
| 11 | Qwen 3 8B (Thinking) | 40.33 |
| 12 | Kimi K2 | 40.05 |
| 13 | Qwen 3 Coder 30B A3B Instruct | 30.79 |
| 14 | Qwen 3 8B (Non-reasoning) | 23.98 |
| 15 | InternLM3-8B-Instruct | 11.44 |
Interactive version: theaggregate.ai/benchmark?slug=optiverse-mathematical-programming · How It Works · Data refreshed daily, snapshot 2026-10-07.