Medical diagnostic reasoning requires verifiable links between visual evidence and clinical knowledge, yet current medical vision language models are aligned with objectives that reward the wrong behavior. Uniform supervised fine tuning lets easy cases dominate updates while clinically critical boundary cases saturate prematurely. Sparse rewards based only on final-answer correctness cannot assess anatomical grounding or clinical logic, inviting reward hacking through answer shortcuts. We propose MedCBR, a two-stage alignment framework coupling boundary-aware preference optimization with rubric-grounded reinforcement learning. First, we construct counterfactual boundary preferences at the model's capability frontier using empirical pass rates and preference pairs that share reasoning prefixes but diverge at decisive diagnostic branches. Difficulty Weighted Boundary Preference Optimization (DBPO) concentrates learning on this calibrated boundary region. Second, we distill reusable clinical rubrics from boundary trajectories and activate them according to prediction entropy. Entropy Gated Grounded Policy Optimization (EGPO) rewards answer correctness, anatomical grounding, and reasoning fidelity while suppressing rubric noise on reliable samples. A reference policy anchored to the boundary-aligned model couples population-level boundary priors with instance-level uncertainty, making diagnostic reasoning more accurate and clinically grounded. Experiments on six medical benchmarks show consistent improvements over strong general, specialized, and reinforcement learning baselines, reaching 61.2% average accuracy with the largest gains on challenging differential diagnosis tasks. Ablations show super-additive gains: combining the two stages exceeds the sum of their individual improvements.
As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanistic analyses remain confined to unimodal text and offer no explanation for why a single misleading sentence can override a correct image based diagnosis, or why a model commits to a confident answer despite insufficient visual evidence. We find that these two safety risks, arbitration failure where textual context overrides visual grounding and brake failure where the model commits without adequate evidence, are mediated by spatially disjoint attention head populations: arbitration heads form a mid-to-deep wideband reflecting cross-layer evidence competition, while brake heads concentrate in a narrow middle-to-late layer band that regulates evidence sufficiency and abstention behavior. To ground these observations in causal circuitry, we introduce CRAFT, which localizes each failure mode to a minimal causal head set via dual criteria and verifies necessity and sufficiency through temporal probes and Tuned Lens trajectory analysis. Excising arbitration heads sharply reduces conflict following with negligible degradation on clean inputs, while excising brake heads restores appropriate abstention under degraded visual evidence. The two interventions target spatially disjoint head sets and produce distinct corrective effects, underscoring the mechanistic separability of the failure modes. Experiments across multiple medical VQA benchmarks and VLM architectures validate both the localization and interventions, demonstrating that the identified heads causally drive each failure mode and that targeted modulation generalises without retraining. The code is available at https://github.com/zhcz328/CRAFT.
We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.