Know2Guess: leaderboard
Know2Guess (2026) tests knowledge-boundary behavior across five domains: models should answer questions with a unique supported answer (zones A to C) and abstain on programmatically built synthetic-unknown questions (zone D). The board is reliability in percent on the 1,080-item test split: correct answers on zones A to C plus abstentions on zone D, over all items; higher is better.
Metric: Reliability (0-1 scaled to %): the share of test items handled correctly, a correct answer on the answer-expected zones A to C and an abstention on the synthetic-unknown zone D (no hidden gold answer); standard prompt on the 1,080-item test split, greedy decoding with at most 128 or 160 output tokens, no retrieval or tools; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 9B (Non-reasoning) | 57.04 | #803 (Qwen 3.5 9B) |
| 2 | Mistral Small 3.2 | 53.98 | #1591 |
| 3 | Qwen 3 8B (Non-reasoning) | 47.22 | #1830 (Qwen 3 8B) |
| 4 | Qwen 2.5 3B Instruct | 38.15 | #5150 |
| 5 | Llama 3 8B Instruct | 35.46 | #4612 |
| 6 | Qwen 2.5 1.5B Instruct | 15.74 | #7470 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=know2guess · How It Works · Data refreshed daily, snapshot 2026-11-13.