AGC-Bench - grapheval_iclr: leaderboard
The "grapheval_iclr" 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: Around September 2027. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Flash (02-23) | 2.04 | #185 |
| 2 | Claude Opus 4.7 | 1.9 | #49 |
| 3 | NVIDIA Nemotron Nano 9B V2 | 1.73 | #823 |
| 4 | GPT-5.4 | 1.41 | #79 |
| 5 | Gemma 4 31B (IT) | 1.34 | #176 |
| 6 | GPT-5.5 | 0.9 | #31 |
| 7 | GPT-4.1 | 0.9 | #248 |
| 8 | Claude Sonnet 4.5 | 0.84 | #142 |
| 9 | GLM-5 | 0.84 | #141 |
| 10 | Claude Opus 4.6 | 0.83 | #63 |
| 11 | GLM-4.6 | 0.82 | #252 |
| 12 | Claude Sonnet 4 | 0.77 | #199 |
| 13 | Kimi K2.5 | 0.74 | #144 |
| 14 | Nemotron Nano 12B v2 VL | 0.74 | #752 |
| 15 | Claude Sonnet 4.6 | 0.71 | #90 |
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-grapheval-iclr · How It Works · Data refreshed daily, snapshot 2026-10-20.