InteractBench - Medium - pass@1: leaderboard

Metric: pass@1 (%; n=10 samples per task). Source: arxiv.org. 16 models tracked.

Top models

#ModelScore
1GPT-5.2 (Thinking)53
2Gemini 3 Pro (Preview)51.2
3DeepSeek V3.2 (Thinking)27.6
4Gemini 2.5 Pro25.9
5GPT-OSS-120B19.9
6Gemini 2.5 Flash14.4
7Claude Opus 4.5 (Thinking)14.1
8DeepSeek V3.2 (Non-reasoning)12.3
9Qwen 3 30B A3B 2507 (Thinking)10.4
10GPT-OSS-20B9.6
11Qwen 3 32B (Thinking)6.5
12Qwen 3 14B (Reasoning)4
13GPT-4.13.9
14DeepSeek R1 Distill Llama 70B3.5
15Qwen 3 32B (Non-reasoning)1.7

Interactive version: theaggregate.ai/benchmark?slug=interactbench-medium-pass-1 · How It Works · Data refreshed daily, snapshot 2026-09-19.