AGC-Bench - grapheval_ai_researcher: leaderboard
The "grapheval_ai_researcher" 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 | GLM-5 | 1.88 | #141 |
| 2 | Nemotron 3 Super | 1.42 | #335 |
| 3 | Gemma 4 26B A4B (IT) | 1.31 | #297 |
| 4 | Nova 2.0 Lite (Non-reasoning) | 1.11 | #417 (Nova 2.0 Lite) |
| 5 | Gemini 2.5 Flash | 1.08 | #245 |
| 6 | Claude Haiku 4.5 | 1.06 | #280 |
| 7 | ERNIE 4.5 300B A47B | 1.02 | #458 |
| 8 | Qwen Turbo | 0.97 | #699 |
| 9 | GLM-4.7 | 0.91 | #192 |
| 10 | Nemotron Nano 12B v2 VL | 0.87 | #752 |
| 11 | Mixtral 8x22B Instruct | 0.81 | #728 |
| 12 | DeepSeek V3.1 Terminus | 0.78 | #220 |
| 13 | GPT-5.5 | 0.73 | #31 |
| 14 | Claude Sonnet 4 | 0.72 | #199 |
| 15 | DeepSeek V3 Chat | 0.7 | #447 |
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-ai-researcher · How It Works · Data refreshed daily, snapshot 2026-10-20.