InteractBench - Easy - pass@5: leaderboard
Metric: pass@5 (%; n=10 samples per task). Source: arxiv.org. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 96.1 |
| 2 | GPT-5.2 (Thinking) | 92.1 |
| 3 | DeepSeek V3.2 (Thinking) | 81.6 |
| 4 | Gemini 2.5 Pro | 76.3 |
| 5 | GPT-OSS-120B | 75 |
| 6 | Claude Opus 4.5 (Thinking) | 59.2 |
| 7 | GPT-OSS-20B | 57.9 |
| 8 | Gemini 2.5 Flash | 56.6 |
| 9 | DeepSeek V3.2 (Non-reasoning) | 50 |
| 10 | Qwen 3 32B (Thinking) | 42.1 |
| 11 | Qwen 3 30B A3B 2507 (Thinking) | 40.8 |
| 12 | GPT-4.1 | 34.2 |
| 13 | Qwen 3 14B (Reasoning) | 26.3 |
| 14 | DeepSeek R1 Distill Llama 70B | 22.4 |
| 15 | Qwen 3 14B (Non-reasoning) | 9.2 |
Interactive version: theaggregate.ai/benchmark?slug=interactbench-easy-pass-5 · How It Works · Data refreshed daily, snapshot 2026-09-19.