LogicScore - HotpotQA Completeness: leaderboard
LogicScore (Yan et al., 2026) Completeness 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. Completeness is the share of answers whose propositions connect the gold answer back to a question entity through shared entities. Percent, higher is better; single run.
Metric: Completeness (%; answers whose extracted propositions link the gold answer back to a question entity; 1,939 HotpotQA distractor-setting items). Source: arxiv.org. Saturation forecast: Around February 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 78.65 | #135 |
| 2 | Qwen 3 30B A3B | 77.09 | #1124 |
| 3 | DeepSeek V3.2 | 76.34 | #403 |
| 4 | Qwen 3 4B | 75.58 | #2297 |
| 5 | GPT-4o | 75.57 | #741 |
| 6 | Qwen 3 8B | 75.36 | #1838 |
| 7 | DeepSeek R1 | 73.91 | #467 |
| 8 | Qwen 3 235B A22B | 73.8 | #622 |
| 9 | O3 | 73.24 | #224 |
| 10 | Qwen 3 14B | 73.18 | #1235 |
| 11 | GPT-5.1 | 72.61 | #225 |
| 12 | Qwen 3 1.7B | 66.17 | #5551 |
| 13 | Qwen 3 0.6B | 52.16 | #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-completeness · How It Works · Data refreshed daily, snapshot 2026-11-12.