Tsinghua and Amap propose OmniVerifier-C1, a graded verifier beating binary judges
Bowei Liu · hf · 2026-10-16
Researchers from Tsinghua University, Amap (Alibaba Group), and Peking University introduce OmniVerifier-C1, a decomposed graded visual verifier for rule-governed generative world modeling.
- Moves beyond binary True/False verifiers: automatically decomposes complex prompts or environment rules into fine-grained components and assigns pointwise scores, giving continuous and more informative critique signals
- Proposes a Decoupled RL training strategy that jointly optimizes sub-question-level binary judgment and global alignment scoring, using precise local feedback to build global fine-grained verification
- Structured, interpretable outputs improve generalization to out-of-domain multimodal critique tasks and enhance structured world-modeling accuracy via pairwise verification
- Code: github.com/Cominclip/OmniVerifier
More from Research
- Kornia releases open-source Rust visual-inertial SLAM pipeline kornia-slam v0.1 — edgarriba · 2026-10-16
- Interpretability debate: one interpretable polysemantic neuron doesn't prove all are — aran_nayebi · 2026-10-16
- Optuna creator Akiba: when 2 hyperparameters cost $20M, what's left for BO/HPO — Stefania_druga · 2026-10-16
- AI will be superhuman at science regardless, and Europe can't opt out — teortaxesTex · 2026-10-16
- SERO augments long-horizon agentic RL with executable oracle continuations — Shanghai-AI-Laboratory · 2026-10-16
- Alibaba open-sources MAI-Gym, an all-in-one GUI agent eval and RL training platform — Tongyi-MAI · 2026-10-16