PodBench - Instruction Following: leaderboard
Metric: Instruction Following (0-100; judge-derived checklist, items scored 0/0.5/1). Source: arxiv.org. Saturation forecast: Around October 2026. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.5 | 96.58 |
| 2 | Gemini 3 Pro (Preview) | 96.09 |
| 3 | Qwen 3 235B A22B (Non-reasoning) | 93.64 |
| 4 | DeepSeek V3 (0324) | 92.71 |
| 5 | Qwen 3 32B (Non-reasoning) | 90.69 |
| 6 | GPT-4o | 89.43 |
| 7 | Qwen 3 14B (Non-reasoning) | 87.07 |
| 8 | Qwen 3 30B A3B (Non-reasoning) | 86.5 |
| 9 | Qwen 2.5 72B Instruct | 83.31 |
| 10 | Qwen 3 8B (Non-reasoning) | 80.79 |
| 11 | Qwen 2.5 32B Instruct | 77.84 |
| 12 | Qwen 3 4B (Non-reasoning) | 76.73 |
| 13 | InternLM3-8B-Instruct | 73.95 |
| 14 | Llama 4 Scout Instruct | 70.14 |
| 15 | Qwen 2.5 7B Instruct | 69.13 |
Interactive version: theaggregate.ai/benchmark?slug=podbench-instruction-following · How It Works · Data refreshed daily, snapshot 2026-09-25.