AGC-Bench - proparalogy: leaderboard
The "proparalogy" subtask of AGC-Bench. A task from AGC-Bench, evaluating models on open-ended creative and figurative-language generation — humor, metaphor, analogy, storytelling and brainstorming — typically graded by LLM-as-judge.
Metric: Dataset z-score. Source: huggingface.co. Saturation forecast: Not forecast. 82 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Flash (02-23) | 0.95 | #185 |
| 2 | Hermes 4 405B | 0.94 | #334 |
| 3 | GPT-5.5 | 0.9 | #31 |
| 4 | Olmo 3.1 32B Instruct | 0.87 | #763 |
| 5 | pixtral-large-2411 | 0.87 | #478 |
| 6 | Kimi K2.5 | 0.86 | #144 |
| 7 | Hermes-3-Llama-3.1-405B | 0.86 | #510 |
| 8 | Claude Sonnet 4.6 | 0.86 | #90 |
| 9 | Claude Sonnet 4.5 | 0.86 | #142 |
| 10 | GPT-5.4 | 0.82 | #79 |
| 11 | Qwen Turbo | 0.79 | #699 |
| 12 | GPT-5.1 | 0.78 | #135 |
| 13 | GPT-4.1 | 0.78 | #248 |
| 14 | Nova Lite (v1) | 0.78 | #830 |
| 15 | Mistral Medium 3.1 | 0.77 | #408 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=agc-bench-proparalogy · How It Works · Data refreshed daily, snapshot 2026-10-20.