CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal Reasoning

TL;DR

CARE introduces a novel failure-centric framework for multimodal reasoning that leverages errors as supervision, significantly improving accuracy and training stability on visual-reasoning benchmarks.

Contribution

The paper proposes CARE, a new method combining contrastive objectives and self-repair to enhance learning from failures in multimodal reasoning tasks.

Findings

Improves accuracy by 4.6 points on Qwen2.5-VL-7B benchmarks.

Achieves state-of-the-art results on MathVista and MMMU-Pro.

Enhances training smoothness and learning from failures.

Abstract

Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures. When all rollouts are wrong, gradients stall; when one happens to be correct, the update usually ignores why the others are close-but-wrong, and credit can be misassigned to spurious chains. We present CARE (Contrastive Anchored REflection), a failure-centric post-training framework for multimodal reasoning that turns errors into supervision. CARE combines: (i) an anchored-contrastive objective that forms a compact subgroup around the best rollout and a set of semantically proximate hard negatives, performs within-subgroup z-score normalization with negative-only scaling, and includes an all-negative rescue to prevent zero-signal batches; and (ii) Reflection-Guided Resampling (RGR), a one-shot structured self-repair that rewrites a representative failure and re-scores it with the same verifier, converting near-misses into usable positives without any test-time reflection. CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures. On Qwen2.5-VL-7B, CARE lifts macro-averaged accuracy by 4.6 points over GRPO across six verifiable visual-reasoning benchmarks; with Qwen3-VL-8B it reaches competitive or state-of-the-art results on MathVista and MMMU-Pro under an identical evaluation protocol.