OptiVerse: leaderboard
Metric: Accuracy (%) over all 1,000 OptiVerse problems; 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 | 55.9 |
| 2 | Gemini 3 Flash | 55.5 |
| 3 | GPT-5.2 (Thinking) | 55.2 |
| 4 | O4 Mini | 51.2 |
| 5 | O3 | 51.1 |
| 6 | Claude Sonnet 4.5 (Thinking) | 49.7 |
| 7 | Gemini 2.5 Pro | 49.6 |
| 8 | DeepSeek V3.2 (Thinking) | 49.5 |
| 9 | Gemini 2.5 Flash | 47.4 |
| 10 | DeepSeek V3.2 (Non-reasoning) | 45.3 |
| 11 | GPT-OSS-120B | 44.9 |
| 12 | Qwen 3 8B (Thinking) | 39.9 |
| 13 | Kimi K2 | 38.9 |
| 14 | Qwen 3 Coder 30B A3B Instruct | 26.1 |
| 15 | Qwen 3 8B (Non-reasoning) | 20 |
Interactive version: theaggregate.ai/benchmark?slug=optiverse · How It Works · Data refreshed daily, snapshot 2026-10-07.