Sahil Mohan Bansal
June 04, 2026
4 min read
June 04, 2026
4 min read
Cut code review time & bugs by 50%
Most installed AI app on GitHub and GitLab
Free 14-day trial
TL;DR: NVIDIA Nemotron 3 Ultra delivers accurate and fast throughput in CodeRabbit's self-hosted AI code reviews.
We are excited to share that CodeRabbit is expanding its support for the NVIDIA Nemotron family of open models, expanding to include Nemotron 3 Super and Nemotron 3 Ultra for AI code review workflows.
Nemotron 3 Super helps with context gathering and summarization whereas Nemotron 3 Ultra helps generate code review comments for many reviews outside of the most complex tiers. This expanded support is available for CodeRabbit's self-hosted customers running its container image on their own infrastructure.
Initial eval results indicate that Nemotron 3 Ultra aligns with our current frontier model ensemble for junior-tier engineering assessments, with similar token efficiency while achieving approximately 2x faster response times. OpenAI and Anthropic models remain the primary engines for producing most of the review comments delivered to your Pull Requests.
Previously we had announced our support of Nemotron 3 Nano and Super, where we reported that a blend of open and frontier models allows us to improve the overall speed of context gathering and PR summarization. This blend of open and frontier models is also more cost efficient by routing different parts of the review workflow to the appropriate model family - PR Summarization with Nemotron and review comments with frontier LLMs.
As with the rest of the Nemotron family, NVIDIA is releasing Ultra as a truly open model, with the weights, training data, and training recipe published alongside it. That openness is part of why Nemotron has been a good fit for self-hosted teams that need to run reviews inside their own environment.
With the support for Nemotron 3 Nano, Super and now Ultra, we can use Nemotron open models for context gathering, PR summarization, and some aspects of review comment generation.
When you open a Pull Request (PR), CodeRabbit’s code review workflow is triggered starting with an isolated and secure sandbox environment where CodeRabbit analyzes code from a clone of the repo. In parallel, CodeRabbit pulls in context signals from several sources:
A lot of this context, along with the code diff being analyzed, is used to generate a PR Summary before any review comments are generated. Summarization is at the heart of every code review and is the key to delivering high signal-to-noise in the review comments. We continue to support Nemotron 3 Nano and Super for the repetitive work of context processing during review summarization, which is critical for our code reviews.
We compared Nemotron 3 Ultra against equivalent frontier models and our analysis found that at the junior reviewer tier Nemotron 3 Ultra:
These results held for both trivial and junior-level review comments. These are early, encouraging results in a specific place: faster, lower-effort reviews where an efficient open model can carry more of the load.
For customers this means faster PR summarization, context gathering and faster code reviews without compromising quality. We are also delighted to support the announcement from NVIDIA today about the expansion of its Nemotron family of open models and are excited to work with the company to help accelerate AI coding adoption across every industry.
Get in touch with the CodeRabbit team to access CodeRabbit’s container image if you would like to run AI code reviews on your self-hosted infrastructure.