PodBench - Thinking Mode - Instruction Following: leaderboard
PodBench's instruction-following stage (Claude-4.5-Opus checks a requirement checklist it derives from each instruction, items scored 0, 0.5 or 1, normalized to 0-100, higher is better) for open models run with reasoning enabled, with the reasoning trace removed before judging. Tencent, ACL 2026.
Metric: Instruction Following (0-100; judge-derived checklist, items scored 0/0.5/1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B (Thinking) | 94.32 | #622 (Qwen 3 235B A22B) |
| 2 | DeepSeek R1 0528 | 94.27 | #425 |
| 3 | Qwen 3 32B (Thinking) | 93.78 | #960 (Qwen 3 32B) |
| 4 | Qwen 3 14B (Reasoning) | 91.79 | #1235 (Qwen 3 14B) |
| 5 | Qwen 3 30B A3B (Thinking) | 89.88 | #1124 (Qwen 3 30B A3B) |
| 6 | Qwen 3 8B (Thinking) | 88.45 | #1838 (Qwen 3 8B) |
| 7 | DeepSeek R1 0528 Qwen3 8B | 85.67 | #2101 |
| 8 | Qwen 3 4B (Reasoning) | 83.04 | #2297 (Qwen 3 4B) |
| 9 | DeepSeek R1 Distill Qwen 32B | 80.12 | #1780 |
| 10 | DeepSeek R1 Distill Llama 8B | 66.94 | #6773 |
| 11 | Qwen 3 1.7B (Thinking) | 62.65 | #5551 (Qwen 3 1.7B) |
| 12 | DeepSeek-R1-Distill-Qwen-7B | 42.59 | #8045 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=podbench-thinking-mode-instruction-following · How It Works · Data refreshed daily, snapshot 2026-11-12.