ProLLM - StackUnseen — leaderboard
ProLLM benchmark evaluating LLMs on unseen Stack Overflow coding questions not in training data. Measures acceptance rate on real developer problems.
Metric: Score (%). Source: www.prollm.ai. Status: saturated. 35 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 | 88.6 |
| 2 | Grok 4 | 88.6 |
| 3 | Gemini 3 Pro (Preview) | 86.2 |
| 4 | Gemini 3 Flash | 83 |
| 5 | GPT-5 Mini | 82.4 |
| 6 | Qwen 3.5 397B A17B | 76.3 |
| 7 | Kimi K2 (Thinking) | 76.1 |
| 8 | MiniMax-M2.5 | 66 |
| 9 | Kimi K2.5 | 64.9 |
| 10 | GPT-5 Nano | 60.4 |
| 11 | GLM-5 | 55.1 |
| 12 | DeepSeek R1 0528 | 52.4 |
| 13 | Mistral Large 3 | 51.6 |
| 14 | DeepSeek V3.1 | 48.1 |
| 15 | Qwen 3 32B | 45.7 |
Interactive version: theaggregate.ai/benchmark?slug=prollm-stackunseen · How the rankings work · Data refreshed daily, snapshot 2026-07-22.