InteractBench - Easy - pass@1: leaderboard
Metric: pass@1 (%; n=10 samples per task). Source: arxiv.org. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 80.3 |
| 2 | GPT-5.2 (Thinking) | 74.2 |
| 3 | DeepSeek V3.2 (Thinking) | 64.5 |
| 4 | Gemini 2.5 Pro | 58.7 |
| 5 | GPT-OSS-120B | 52.1 |
| 6 | Gemini 2.5 Flash | 33.2 |
| 7 | Claude Opus 4.5 (Thinking) | 31.8 |
| 8 | GPT-OSS-20B | 31.3 |
| 9 | DeepSeek V3.2 (Non-reasoning) | 30.8 |
| 10 | Qwen 3 32B (Thinking) | 23.7 |
| 11 | Qwen 3 30B A3B 2507 (Thinking) | 19 |
| 12 | GPT-4.1 | 16.3 |
| 13 | Qwen 3 14B (Reasoning) | 14.7 |
| 14 | DeepSeek R1 Distill Llama 70B | 10.5 |
| 15 | Qwen 3 32B (Non-reasoning) | 3.2 |
Interactive version: theaggregate.ai/benchmark?slug=interactbench-easy-pass-1 · How It Works · Data refreshed daily, snapshot 2026-09-19.