AGC-Bench - speak_to_structure: leaderboard
The "speak_to_structure" 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 March 2027. 82 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 1.51 | #83 |
| 2 | GPT-5.5 | 1.49 | #31 |
| 3 | Claude Opus 4.7 | 1.47 | #49 |
| 4 | GPT-5.4 | 1.32 | #79 |
| 5 | Kimi K2 0905 | 1.15 | #288 |
| 6 | Claude Sonnet 4.5 | 1.13 | #142 |
| 7 | Kimi K2.5 | 1.07 | #144 |
| 8 | Claude Opus 4.6 | 1.05 | #63 |
| 9 | Claude Sonnet 4 | 1.05 | #199 |
| 10 | Kimi K2.6 | 1.03 | #104 |
| 11 | GLM-5 | 0.89 | #141 |
| 12 | GPT-5.4 Mini | 0.82 | #212 |
| 13 | GLM-5.1 | 0.8 | #97 |
| 14 | Claude 3.7 Sonnet | 0.79 | #249 |
| 15 | DeepSeek V3.1 Terminus | 0.77 | #220 |
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-speak-to-structure · How It Works · Data refreshed daily, snapshot 2026-10-20.