Soft Adaptive Policy Optimization

CodeV has been introduced as a code-based visual agent that utilizes Tool-Aware Policy Optimization (TAPO) to enhance visual reasoning in AI models. This development highlights the need for faithful visual reasoning, as existing models often achieve high accuracy while misusing visual tools or ignoring relevant outputs. The proposed faithfulness evaluation protocol aims to address these shortcomings by measuring the relevance of intermediate visual tool outputs.

The Adversarial Confusion Attack has been introduced as a new threat to multimodal large language models (MLLMs), aiming to disrupt their output by generating incoherent or confidently incorrect responses. This attack utilizes adversarial images to compromise the reliability of MLLM-powered agents, demonstrating its effectiveness across various models, including proprietary ones like GPT-5.1.

A systematic analysis of test-time thinking in large vision language models (LVLMs) has been conducted, revealing that generating explicit reasoning chains can enhance performance, but longer chains may lead to errors that detract from visual reasoning. The study evaluated ten variants from the InternVL3.5 and Qwen3-VL families on the MMMU-val dataset under generous token budgets and multi-pass decoding.

The introduction of the Discriminative Constrained Optimization (DisCO) framework aims to enhance large reasoning models (LRMs) by addressing limitations found in the Group Relative Policy Optimization (GRPO) method, particularly regarding question-level difficulty bias. DisCO emphasizes a discriminative objective and utilizes non-clipping reinforcement learning surrogate objectives, marking a significant shift in reinforcement learning strategies for LRMs.

Recent research highlights the challenges of ensuring honesty in language models during deductive reasoning tasks, where models must derive conclusions strictly from given premises. The study introduces a framework for honest deductive reasoning, emphasizing the need for models to abstain from answering when conclusions are not logically entailed by the premises.

ExPO-HM (Explain-then-Detect Policy Optimization for Hateful Memes) has been proposed to enhance the detection of hateful memes, addressing limitations in existing models that primarily provide binary predictions without context. This new approach aims to incorporate reasoning similar to human annotators, improving the understanding of policy-relevant cues such as targets and attack types.

A new framework called ReVeL (Rewrite and Verify by LLM) has been proposed to enhance the multiple-choice question answering (MCQA) format used in evaluating multimodal language models. This framework transforms MCQA into open-form questions while ensuring answers remain verifiable, addressing issues of answer guessing and unreliable accuracy metrics during reinforcement fine-tuning (RFT).

A new study introduces Periodic Asynchrony as a method to enhance on-policy reinforcement learning, addressing the inefficiencies of synchronous execution in mainstream frameworks. By separating inference and training, this approach allows for independent scaling of components while maintaining accuracy equivalent to traditional methods.