DR Tulu: open 8B deep-research model with evolving rubrics matches OpenAI DR
AkariAsai · x · 2026-09-05
Akari Asai highlights two open research efforts:
- DR Tulu (ICML oral): introduces RLER, reinforcement learning with evolving rubrics that co-evolve with the policy model during training, incorporating newly explored search results and contrasting responses. The resulting DR Tulu-8B is the first fully open model trained for open-ended long-form deep research, beating Tongyi DR by 15.6% on average across four benchmarks and edging out OpenAI DR by 0.7%.
- AgentIR: to be presented at COLM next month.
Author list includes Luke Zettlemoyer, Hannaneh Hajishirzi, Pang Wei Koh and others from UW/AI2.
More from Research
- Experimental boolean-logic LM: microsecond forward pass on one CPU thread, but painfully slow training — cephaloform · 2026-10-23
- SkyRL and Lila Sciences scale GLM-5.3-Flash training to 1M-token context, 2.1x faster at 256K — AndrewLBeam · 2026-10-23
- AdaDex turns seconds of human hand motion into generalizable dexterous robot skills — mangahomanga · 2026-10-23
- Eric Topol: first demonstration that AI cuts diagnostic testing, lowers costs and boosts accuracy — EricTopol · 2026-10-23
- ReAPO paper: conditioning retries on failures beats GRPO in RLVR reasoning training — jasondeanlee · 2026-10-23
- Quantum-JEPA: first JEPA world model deployable on real quantum circuits (QPU) — DrMorganLevine · 2026-10-23