CUDA 12.6 remains the minimum stable version for full RTX 50-series (Blackwell) compatibility. Developers using Blackwell-based GPUs (like the RTX 5090) continue to rely on 12.6 or newer to avoid the compute capability issues found in older CUDA 11.x versions.

As the high-performance computing (HPC) and AI communities wind down a landmark year, NVIDIA has delivered a final, substantial gift to the developer ecosystem. The December 2025 update to the CUDA ecosystem (build 12.6.x) has landed, and while it may not carry the marketing flash of a major version number jump, industry insiders are calling it one of the most critical stability and optimization releases in recent memory.

This update serves as a bookend to the Blackwell era. As the industry looks toward 2026, whispers of the Rubin architecture are already circulating. However, NVIDIA’s messaging is clear: CUDA 12.6 is the stable bedrock upon which the AI factories of 2026 will be built.

It’s been over 15 months since NVIDIA released CUDA 12.6 in September 2024. Despite ongoing rumors in developer forums, — nor will it.

For those operating on the 12.6 update path as of December 2025, the following features define the experience: Nsight Compute Release History - NVIDIA Developer