YapBench — leaderboard
Measures LLM verbosity on brevity-ideal prompts. Models are scored on excess length beyond a minimal baseline answer using YapScore and YapIndex metrics.
Metric: YapIndex (lower is better). Source: huggingface.co. 108 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GLM-4.5 | 1427 |
| 2 | nova-2-lite-v1 | 896.3 |
| 3 | Qwen 3 32B | 730.2 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 666.3 |
| 5 | OLMo 3.1 32B (Thinking) | 563 |
| 6 | Qwen 3 14B | 558 |
| 7 | Gemini 2.5 Pro | 543.8 |
| 8 | Qwen 3 VL 8B Instruct | 518 |
| 9 | Claude Sonnet 5 | 507.7 |
| 10 | Gemini 2.5 Flash Lite | 476.2 |
| 11 | DeepSeek V3.2 (Thinking) | 462.3 |
| 12 | Devstral Small 2 | 449 |
| 13 | Mistral Large 3 | 444 |
| 14 | GLM-4.6V | 443.8 |
| 15 | DeepSeek V3.2 | 426.2 |
Interactive version: theaggregate.ai/benchmark?slug=yapbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.