On 4 June 2026, NVIDIA announced Nemotron 3 Ultra 550B-A55B. NVIDIA released a large hybrid model together with extensive dataset disclosure and reproducible NeMo evaluation recipes.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 4 June 2026 |
| Availability or weight release | 4 June 2026 as BF16, base and NVFP4 checkpoints |
| Release type | open-weight frontier-scale reasoning and agent model |
| Access | OpenMDW 1.1 licensed weights, training recipes and NVIDIA serving paths |
| Architecture | a 550B hybrid LatentMoE with 55B active parameters, Mamba-2, attention and multi-token prediction |
| Maximum stated context | up to one million tokens |
What changed
Nemotron 3 Ultra’s differentiator is the breadth of its release package. NVIDIA disclosed major training datasets, post-training methods, evaluator containers and a low-precision checkpoint, making it easier to investigate how the reported results were produced. The hybrid Mamba-2 and attention design also tests whether very long context can be served more efficiently than a conventional full-attention model at this scale.
The practical comparison is therefore not simply whether Nemotron 3 Ultra 550B-A55B has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| SWE-Bench Verified | 70.7 | Reported repository-fix result |
| Terminal-Bench 2.1 | 56.4 | Shows a weaker area than its reasoning results |
| IMOAnswerBench with tools | 92.3 | Tool-assisted mathematical reasoning result |
These are release-time results, not independently reproduced guarantees. The model defaults to 256K in common servers and needs explicit configuration for one million tokens; several benchmarks still use internal scaffolding pending release. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
Nemotron 3 Ultra 550B-A55B is described as a 550B hybrid LatentMoE with 55B active parameters, Mamba-2, attention and multi-token prediction with up to one million tokens of stated context. Its access position at verification time is OpenMDW 1.1 licensed weights, training recipes and NVIDIA serving paths. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For Nemotron 3 Ultra 550B-A55B, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
Nemotron 3 Ultra is a consequential open-weight systems release. Its raw size limits casual deployment, but the disclosed recipes and evaluations make it valuable for organisations with serious NVIDIA infrastructure.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: NVIDIA official model card via Hugging Face. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
