Updated

NV-Reason-CT

NVIDIA released NV-Reason-CT, an open research vision language model that applies chain of thought reasoning to 3D CT scans. It generates structured reports and supports multi-step follow-up questions for chest and abdominal CT, and is positioned as a foundation model for researchers rather than a cleared clinical product.

By No Code MBA · Facts checked

Provider
NVIDIA
Announcement date
Status at last check
Generally available
Model family
NV-Reason
Preceding model
NV-Reason-CXR
Capabilities
Text, Image

What changed

NV-Reason-CT extends NVIDIA's NV-Reason-CXR chest X-ray approach to 3D CT volumes of the chest and abdomen. It pairs a Qwen3.5-4B language model with a 3D ViT encoder, turning each 192-cubed voxel CT into 13,824 vision tokens, and produces structured reports, chain of thought reasoning and multi-turn answers. NVIDIA trained it on about 550,000 QA examples plus GRPO reinforcement learning, and reports macro-F1 of 0.614 on CT-RATE versus 0.581 for VoxelFM.

Access and availability

Weights on Hugging Face at nvidia/NV-Reason-CT, a demo on Hugging Face Spaces (nvidia/nv-reason-ct), and inference and training code on GitHub at NVIDIA-Medtech/NV-Reason-CT.

Official access documentation

What you can build

You can post-train NV-Reason-CT on the open weights for CT-specific applications, but it is a research foundation model, not approved for clinical diagnostic use.

Limitations

NVIDIA says it is a research and development foundation, not an autonomous diagnostic system or a cleared clinical product, intended for further post-training. Coverage is limited to chest and abdominal CT, and benchmark comparisons are NVIDIA's own.

Sources and release history