Citation and Evidence Recall on NeoQA
92.8RkGPT-5.2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5.2Reasoning condition=Reasoning2025.10 | 92.8 | 96.1 | 93.8 | |
| GPT-5-MiniReasoning condition=Reasoning2025.10 | 89 | 88.1 | 91 | |
| GPT-OSS-120BReasoning condition=Reasoning2025.10 | 88.5 | 86 | 91.1 | |
| GPT-OSS-20BReasoning condition=Reasoning2025.10 | 77.1 | 57.2 | 82.2 | |
| Qwen3-32BReasoning condition=Reasoning2025.10 | 75.9 | 84.8 | 81.4 | |
| Mistr-S-3.2-24BReasoning condition=No Reasoning2025.10 | 75.6 | 71.7 | 73.2 | |
| Qwen3-14BReasoning condition=Reasoning2025.10 | 75.4 | 81.4 | 79.7 | |
| Qwen3-8BReasoning condition=Reasoning2025.10 | 74.5 | 77.4 | 80.8 | |
| Phi-4Reasoning condition=No Reasoning2025.10 | 73.7 | 49.4 | 72.2 | |
| Oracle-BM25Reasoning condition=No Reasoning2025.10 | 70.8 | 100 | 70.8 | |
| Qwen3-4BReasoning condition=Reasoning2025.10 | 70.1 | 67.2 | 75.2 | |
| Oracle-DRAGReasoning condition=No Reasoning2025.10 | 68 | 100 | 68 | |
| Llama-3.1-8BReasoning condition=No Reasoning2025.10 | 63.6 | 55.4 | 61.6 | |
| Phi-4-MiniReasoning condition=No Reasoning2025.10 | 61.8 | 42.5 | 63 | |
| Llama-3.3-70BReasoning condition=No Reasoning2025.10 | 61.5 | 52.8 | 61.4 | |
| Ministr-8BReasoning condition=No Reasoning2025.10 | 51 | 47.6 | 52.5 | |
| Llama-3.2-3BReasoning condition=No Reasoning2025.10 | 47.9 | 40.5 | 48.6 | |
| Qwen3-1.7BReasoning condition=Reasoning2025.10 | 45.5 | 48.3 | 53 | |
| Qwen3-0.6BReasoning condition=Reasoning2025.10 | 27.6 | 37 | 35.7 | |
| Llama-3.2-1BReasoning condition=No Reasoning2025.10 | 8 | 19.1 | 8.6 |