PodBench - Thinking Mode: leaderboard
PodBench score, the mean of Claude-4.5-Opus instruction-following and script-quality judgments (0-100, higher is better), for open models run with reasoning enabled: the hybrid Qwen3 models with thinking switched on, DeepSeek-R1-0528 and the R1 distills. The reasoning trace is removed before judging, and the 800 Chinese and English podcast-script queries are the same as on PodBench; Tencent, ACL 2026.
Metric: PodBench Score (0-100; mean of instruction following and script quality). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 0528 | 77.71 | #425 |
| 2 | Qwen 3 235B A22B (Thinking) | 77.37 | #622 (Qwen 3 235B A22B) |
| 3 | Qwen 3 32B (Thinking) | 76.85 | #960 (Qwen 3 32B) |
| 4 | Qwen 3 14B (Reasoning) | 74.33 | #1235 (Qwen 3 14B) |
| 5 | Qwen 3 30B A3B (Thinking) | 72.33 | #1124 (Qwen 3 30B A3B) |
| 6 | Qwen 3 8B (Thinking) | 71.17 | #1838 (Qwen 3 8B) |
| 7 | DeepSeek R1 0528 Qwen3 8B | 68.49 | #2101 |
| 8 | Qwen 3 4B (Reasoning) | 67.1 | #2297 (Qwen 3 4B) |
| 9 | DeepSeek R1 Distill Qwen 32B | 64.09 | #1780 |
| 10 | DeepSeek R1 Distill Llama 8B | 56.08 | #6773 |
| 11 | Qwen 3 1.7B (Thinking) | 52.47 | #5551 (Qwen 3 1.7B) |
| 12 | DeepSeek-R1-Distill-Qwen-7B | 39.4 | #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 · How It Works · Data refreshed daily, snapshot 2026-11-12.