NVIDIA / language model; inputs not listed

Nemotron 3 Ultra 550B A55B

EVIDENCE AT A GLANCE

Quality, size and local speed

LiveBench 2026-06-25 overall ↗Model card / primary source ↗

LiveBench · overall67.4/ 100

#19 among 22 evaluated open models in this release

Model-weight disk · Q4 estimate~343.8 GBFormula, not a published download size

Planning memory · Q4 estimate~421 GBIncludes planning allowance; not measured VRAM

Local speedNo exact measurementDo not infer speed from quality or memory

Quality · Published benchmark result; release and source linked aboveMemory · Planning estimate based on documented model size; not measured hardware fitLocal speed · No exact local measurement

Where it performs best

LiveBench 2026-06-25 · 0–100 ↗

LiveBench 2026-06-25 measures the named benchmark variant, not a local Q4 run. Scores were collected 2026-09-26; the seven categories share the same 0–100 scale. This is not the Artificial Analysis Intelligence Index. Q4 size covers only the counted weights; multimodal auxiliary components may add files.

SAME BENCHMARK RELEASE

Quality versus estimated Q4 weight size

Full ranking ↗

Reviewed language checkpoints with a LiveBench score and a parameter-based size estimate. The highlighted point is this model when both values are available.

Horizontal axis: estimated Q4 model-weight disk size in decimal GB (log scale). Vertical axis: LiveBench overall / 100. This formula is not the published checkpoint size, runtime memory or a speed test. Models without both inputs are left out of this chart, not scored as zero.

The Nemotron 3 Ultra 550B A55B checkpoint is published by NVIDIA for language and reasoning. Its weights are available under OpenMDW 1.1; that alone does not establish that its code and all components are open source.

When to consider it

Compare Nemotron 3 Ultra 550B A55B with other language and reasoning checkpoints using its linked publisher card and the LiveBench 2026-06-25 snapshot. This is quality evidence, not a local speed result.

Check before choosing

The generic Q4 planning estimate is about 421 GB from 550B stated parameters. It is not a measured minimum; context, batch size, runtime and quantization change actual memory use.

What is documented

Checkpoint / version: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 ↗
Weight license: OpenMDW 1.1 ↗
Task and modality: language and reasoning · language model; inputs not listed
Count and basis: 550B · manufacturer specification · 55B active
Context: 262,144 tokens published for this model
Quality evidence: 67.4 · LiveBench 2026-06-25 overall · source ↗
Runtime and precision: No local execution verified. Check the publisher's supported runtime and checkpoint precision; batch and context memory are not measured here.

Run locally: known and unknown

~421 GB generic Q4 planning memory. Formula: parameters × 0.625 bytes, plus 20% working allowance and 8 GB reserve. This is neither measured VRAM nor checkpoint download size.

Q4 planning estimate only. Check checkpoint format, runtime support, KV cache, context, batch size and offload before choosing hardware.

Evidence level: Generic estimate only; no execution reproduced by this site.

Before downloading

Check the exact checkpoint and license linked here.
Confirm a compatible runtime and precision or quantization for your platform.
Allow for context, batch size and any extra media components beyond the weights.

Catalog reviewed: 2026-09-30. Benchmark release: 2026-06-25; snapshot collected: 2026-09-26.