LogicScore - HotpotQA Determinateness: leaderboard
LogicScore (Yan et al., 2026) Determinateness on 1,939 chain-structured HotpotQA questions (distractor setting): the model writes a cited step-by-step answer and a short answer from the retrieved documents, and GPT-4o-mini splits the long answer into propositions and entity triples. Determinateness is the share of answers whose own short answer o4-mini recovers from the long-form answer alone, a consistency check rather than a correctness check. Percent, higher is better; single run.
Metric: Determinateness (%; answers whose own short answer o4-mini re-derives from the long-form answer alone; 1,939 HotpotQA distractor-setting items). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 99.15 | #225 |
| 2 | O3 | 98.92 | #224 |
| 3 | Qwen 3 235B A22B | 98.87 | #622 |
| 4 | Gemini 3 Pro | 98.56 | #135 |
| 5 | DeepSeek R1 | 98.56 | #467 |
| 6 | DeepSeek V3.2 | 97.96 | #403 |
| 7 | Qwen 3 30B A3B | 97.77 | #1124 |
| 8 | Qwen 3 8B | 97.68 | #1838 |
| 9 | GPT-4o | 97.36 | #741 |
| 10 | Qwen 3 14B | 97.31 | #1235 |
| 11 | Qwen 3 4B | 96.87 | #2297 |
| 12 | Qwen 3 1.7B | 92.9 | #5551 |
| 13 | Qwen 3 0.6B | 81.52 | #8392 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=logicscore-hotpotqa-determinateness · How It Works · Data refreshed daily, snapshot 2026-11-12.