V-Rubrics: Visual Faithfulness via Rubric-Based Reinforcement Learning

Vision-language models can produce fluent answers that are visually wrong: a single unsupported object, chart value, or intermediate inference can invalidate an otherwise plausible response. We argue that this is a credit-assignment failure in multimodal post-training. Scalar outcome rewards say whether an answer is acceptable, but not which visual facts are grounded, which reasoning steps are valid, or which instruction constraints are missed. We introduce Visual Rubrics-Based Reinforcement Learning, which decomposes reference responses into atomic propositions and scores generated answers along Visual Faithfulness (VF), Reasoning Consistency (RC), and Instruction Following (IF). The resulting rubric items provide structured partial credit and localize rubric credit when supporting evidence spans are available. We first obtain an SFT checkpoint by fine-tuning Qwen3-VL-8B-Instruct on the public OpenMMReasoner-SFT-874K corpus, adapting OpenMMReasoner’s cold-start data recipe. We construct V-Rubrics 50K, a 50,248-example training set from 17 visually grounded sources, by applying rule-based filters before deriving example difficulty from rejection-sampling scores and then annotating every example with Gemini-3-Pro under the same structured prompt and protocol. The same SFT checkpoint supplies the rejection-sampling rollouts and initializes two GRPO variants using either scalar answer credit or component-wise, prefix-localized rubric credit. Rubric-based GRPO improves over both the shared SFT baseline and answer-only GRPO, with the largest gains on knowledge-oriented and visually grounded reasoning benchmarks. The results show rubrics as a useful reward abstraction for visual post-training.