Compositional Multi-hop QA on Bamboogle (Success Rate)
77.6Success RateLlama 3.1 405B
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama 3.1 405BScale group=Large Scale Models (> 10B), Model Size=405B2025.12 | 77.6 | |
| Llama 3.1 70BScale group=Large Scale Models (> 10B), Model Size=70B2025.12 | 76.8 | |
| Qwen 3 14BScale group=Large Scale Models (> 10B), Model Size=14B2025.12 | 76 | |
| GenEnvScale group=7B Models, Model Size=7B2025.12 | 76 | |
| Qwen 3 32BScale group=Large Scale Models (> 10B), Model Size=32B2025.12 | 71.2 | |
| Qwen 2.5 72BScale group=Large Scale Models (> 10B), Model Size=72B2025.12 | 69.6 | |
| ReSearchScale group=7B Models, Model Size=7B2025.12 | 68 | |
| Qwen 2.5 7BScale group=7B Models, Model Size=7B2025.12 | 68 | |
| SearchR1Scale group=7B Models, Model Size=7B2025.12 | 67.2 | |
| ToRLScale group=7B Models, Model Size=7B2025.12 | 34.4 | |
| GPT-OSS 20BScale group=Large Scale Models (> 10B), Model Size=20B2025.12 | 33.6 | |
| GPT-OSS 120BScale group=Large Scale Models (> 10B), Model Size=120B2025.12 | 29.6 |