AGC-Bench - scar: leaderboard
The "scar" 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 April 2027. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 1.12 | #83 |
| 2 | GLM-5.1 | 1.08 | #97 |
| 3 | GLM-5-Turbo | 1 | #119 |
| 4 | Gemma 4 31B (IT) | 1 | #176 |
| 5 | DeepSeek V3.2 | 0.97 | #205 |
| 6 | GLM-5 | 0.96 | #141 |
| 7 | GPT-5.5 | 0.95 | #31 |
| 8 | Claude Sonnet 4.6 | 0.93 | #90 |
| 9 | DeepSeek V3.1 NEX N1 | 0.86 | #327 |
| 10 | Claude Sonnet 4 | 0.85 | #199 |
| 11 | DeepSeek V3 (0324) | 0.84 | #340 |
| 12 | GPT-5.4 | 0.83 | #79 |
| 13 | GPT-5.1 | 0.83 | #135 |
| 14 | Claude Opus 4.7 | 0.79 | #49 |
| 15 | Claude Opus 4.6 | 0.75 | #63 |
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-scar · How It Works · Data refreshed daily, snapshot 2026-10-20.