Agentic Reasoning on FRAMES n=50 (full)
77.31AccuracyGPT-5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5Web search=true, Trials per question=642025.12 | 77.31 | — | 0.088 | 0.496 | |
| Claude 4.5 HaikuWeb search=true, Trials per question=642025.12 | 68.37 | — | 0.144 | 0.655 | |
| Claude 4.5 SonnetWeb search=true, Trials per question=642025.12 | 66.44 | — | 0.156 | 0.689 | |
| GPT-4oWeb search=true, Trials per question=642025.12 | 63.54 | — | 0.174 | 0.735 | |
| Gemini 2.5 ProWeb search=false, Trials per question=642025.12 | 62.34 | — | 0.174 | 0.713 | |
| Deepseek-v3p1Web search=false, Trials per question=642025.12 | 44.75 | — | 0.157 | 0.663 | |
| GPT-4oWeb search=false, Trials per question=642025.12 | 38.16 | — | 0.171 | 0.712 | |
| Qwen3-235b-a22bWeb search=false, Trials per question=642025.12 | 34.22 | — | 0.169 | 0.617 |