StakeBench - Revealed Side: leaderboard
Metric: Strict accuracy (%) on G2: which side (YES or NO) a positioned commenter holds, abstaining counted as wrong, Polymarket and Manifold comments of resolved markets, macro-average over 18 topic and platform splits, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around 2033. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.2 3B Instruct | 53.4 |
| 2 | GPT-5.5 | 44.8 |
| 3 | Qwen 3 32B (Non-reasoning) | 43 |
| 4 | Qwen 3 14B (Non-reasoning) | 42.3 |
| 5 | Gemma 2 9B | 39.7 |
| 6 | Gemini 2.5 Flash | 38.4 |
| 7 | Claude Haiku 4.5 | 38.1 |
| 8 | DeepSeek R1 Distill Llama 8B | 34.5 |
| 9 | Qwen 3 30B A3B (Non-reasoning) | 34.4 |
| 10 | Mistral 7B Instruct (v0.3) | 33.5 |
| 11 | Qwen 3 8B (Non-reasoning) | 32.5 |
Interactive version: theaggregate.ai/benchmark?slug=stakebench-revealed-side · How It Works · Data refreshed daily, snapshot 2026-10-07.