AGC-Bench - irfl: leaderboard
The "irfl" 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. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 1.59 | #79 |
| 2 | GPT-5.5 | 1.53 | #31 |
| 3 | GPT-4o | 1.3 | #343 |
| 4 | ERNIE-4.5-VL-424B-A47B | 1.21 | #391 |
| 5 | GPT-4o Mini | 1.11 | #599 |
| 6 | Gemma 3 12B (IT) | 0.89 | #664 |
| 7 | Mistral Small 3.1 | 0.8 | #610 |
| 8 | Llama 4 Scout | 0.69 | #656 |
| 9 | Claude Sonnet 4.6 | 0.58 | #90 |
| 10 | Qwen 3.5 Flash (02-23) | 0.58 | #185 |
| 11 | Gemma 4 26B A4B (IT) | 0.57 | #297 |
| 12 | Claude Opus 4.5 | 0.54 | #83 |
| 13 | Mistral Large 3 | 0.49 | #401 |
| 14 | Claude 3.7 Sonnet | 0.33 | #249 |
| 15 | Gemma 3 27B (IT) | 0.29 | #523 |
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-irfl · How It Works · Data refreshed daily, snapshot 2026-10-20.