Deep research agent evaluation on GAIA
71.3GAIA ScoreLiteRes.
Evaluation Results
| Method | Links | |
|---|---|---|
| LiteRes.Model=4B, RL Alg.=GRPO, Env.=Static, Reward=Outcome, Arch.=Single, Reflect.=No, Adv. Tr.=No, Cost/Step=$02026.06 | 71.3 | |
| DeepRubricModel=8B, RL Alg.=Rubric-GRPO, Env.=Static, Reward=Rubric, Arch.=Single, Reflect.=Partial, Adv. Tr.=No, Cost/Step=$02026.06 | 69.8 | |
| CaRRModel=7B, RL Alg.=C-GRPO, Env.=Static, Reward=Cit.-Aware, Arch.=Single, Reflect.=No, Adv. Tr.=No, Cost/Step=$0.01–.052026.06 | 65.2 | |
| S-R1++Model=7B/32B, RL Alg.=REINFORCE, Env.=Live-Web, Reward=Outcome, Arch.=Single, Reflect.=No, Adv. Tr.=No, Cost/Step=$0.01–.052026.06 | 63.1 |