AGC-Bench - STEM: leaderboard
The "STEM" 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: JRT z-score. Source: huggingface.co. Saturation forecast: Around March 2027. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 0.81 | #79 |
| 2 | GLM-5 | 0.76 | #141 |
| 3 | Gemma 4 31B (IT) | 0.73 | #176 |
| 4 | Qwen 3.5 Flash (02-23) | 0.67 | #185 |
| 5 | GPT-5.5 | 0.65 | #31 |
| 6 | GLM-5.1 | 0.65 | #97 |
| 7 | GLM-5-Turbo | 0.58 | #119 |
| 8 | Kimi K2 0905 | 0.58 | #288 |
| 9 | Gemma 4 26B A4B (IT) | 0.57 | #297 |
| 10 | Gemini 2.5 Flash | 0.56 | #245 |
| 11 | DeepSeek V3.1 NEX N1 | 0.54 | #327 |
| 12 | ERNIE 4.5 300B A47B | 0.53 | #458 |
| 13 | Claude Opus 4.7 | 0.53 | #49 |
| 14 | Nemotron 3 Super | 0.53 | #335 |
| 15 | GLM-4.7 | 0.51 | #192 |
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-stem · How It Works · Data refreshed daily, snapshot 2026-10-20.