Absurd World - Few-Shot - REAL: leaderboard
Absurd World (Albright et al., 2026) few-shot Determine Outcome task: from a text commentary of a five-shot-per-team penalty shootout under the REAL ruleset, which mirrors ordinary shootout scoring (a hit scores, the higher total wins), the model names the winner after three solved examples of the same ruleset, with questions drawn from the zero-shot prompts that at most the median number of models answered correctly. Accuracy over 100 prompts per model, in percent; higher is better.
Metric: Determine-Outcome Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 Nemotron Ultra 253B v1 | 100 | #1027 |
| 2 | Gemini 2.5 Flash | 99 | #452 |
| 3 | DeepSeek R1 0528 | 98 | #425 |
| 4 | Qwen 2.5 72B Instruct | 97 | #1093 |
| 5 | GPT-4.1 Mini | 96 | #775 |
| 6 | GPT-4o Mini | 92 | #1627 |
| 7 | Llama 3.1 405B Instruct | 92 | #1114 |
| 8 | Gemini 2.0 Flash (001) | 92 | #724 |
| 9 | Claude 3.5 Haiku | 83 | #1563 |
| 10 | Hermes-3-Llama-3.1-405B | 83 | #1022 |
| 11 | Llama 4 Maverick Instruct | 79 | #1125 |
| 12 | Gemma 2 27B (IT) | 37 | #1815 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=absurd-world-few-shot-real · How It Works · Data refreshed daily, snapshot 2026-11-12.