SPLIT-RL: Staged Perception-Language Reasoning Training with Claim-Level Advantages

SPLIT-RL, a staged post-training approach that trains visual reasoning and language reasoning in disjoint phases, improves average accuracy over GRPO by 1.4 to 6.1 points across Qwen3-VL models from 2B to 30B-A3B and InternVL3.5-8B, according to the arXiv paper 2610.10889v1. The method also introduces Claim-Level Advantage (CLA-GRPO), which decomposes visual-reasoning-phase rollouts into atomic visual claims and assigns fine-grained advantage at the claim level based on visual-type group formation. An oracle-based diagnostic showed answer-only GRPO leaves perception unchanged, whereas SPLIT-RL improves both visual and language reasoning, with the trained policy evaluated using a single chain-of-thought call at inference time.

arXiv:2610.10889v1 Announce Type: new Abstract: Vision-Language VL reasoning requires a model to both extract relevant and accurate information from an image visual reasoning, VR , and to infer the answer from it language reasoning, LR . Reinforcement learning with verifiable rewards typically trains both through a single chain-of-thought with a final-answer reward. This gives every CoT token the same sequence-level advantage, failing to distinguish capability specific errors. We propose SPLIT-RL, a staged post-training approach that trains VR and LR in disjoint phases. Because a group's rollouts differ along one capability at a time, the group-relative advantage isolates it, and each phase is optimized using phase-specific reward. We further introduce Claim-Level Advantage CLA-GRPO , which decomposes VR-phase rollouts into atomic visual claims and provides a fine-grained advantage at claim level based on visual-type group formation. Although trained in two phases, trained policy is evaluated like GRPO model, with a single CoT call at inference time. Under this protocol, SPLIT-RL improves average accuracy over GRPO by 1.4-6.1 points across Qwen3-VL models from 2B to 30B-A3B and InternVL3.5-8B. Evaluating each capability using an oracle based diagnostic shows that answer-only GRPO leaves perception unchanged, whereas SPLIT-RL improves both VR and LR.