COIN-Bench - Correctness: leaderboard
Metric: Correctness (%): share of the questionnaire's inferred consensus statements that a retrieval-augmented verifier (COIN-RAG, TF-IDF plus MiniLM retrieval over the raw comments) finds consistent with the majority view in the discussions; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 90.06 | #93 |
| 2 | GPT-5.2 | 85.66 | #105 |
| 3 | O3 | 80.35 | #121 |
| 4 | GPT-4o | 75.75 | #333 |
| 5 | Qwen 2.5 72B Instruct | 64.11 | #436 |
| 6 | GPT-5 | 62.65 | #91 |
| 7 | Qwen 3 30B A3B | 61.6 | #488 |
| 8 | Qwen 2.5 14B Instruct | 60.88 | #634 |
| 9 | GPT-4.1 | 59.05 | #240 |
| 10 | DeepSeek R1 Distill Qwen 14B | 58.45 | #828 |
| 11 | Qwen 3 32B | 55.26 | #424 |
| 12 | Qwen 2.5 32B Instruct | 54.95 | #491 |
| 13 | DeepSeek R1 Distill Qwen 32B | 53.9 | #640 |
| 14 | Claude 3.5 Sonnet | 53.35 | #337 |
| 15 | Llama 3.1 8B Instruct | 52.67 | #1018 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=coin-bench-correctness · How It Works · Data refreshed daily, snapshot 2026-10-11.