GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

NVIDIA Nemotron 3 Ultra Claims US Open-Weights Lead at 550B Params, Intelligence Index 48

NVIDIA’s Nemotron 3 Ultra, announced in Jensen Huang’s Computex keynote, takes the top spot among US-origin open-weights models with an Artificial Analysis Intelligence Index score of 48. That puts it 9 points ahead of Google’s Gemma 4 31B (39), 12 above Nemotron 3 Super (36), and 15 above gpt-oss-120b (33). Artificial Analysis evaluated the model directly in partnership with NVIDIA.

Architecture and Scale

At approximately 550 billion total parameters with 90% sparsity, Nemotron 3 Ultra activates 55B parameters per forward pass. It is the largest Nemotron 3 release by a significant margin. NVIDIA will ship it in NVFP4 quantization alongside BF16 weights, consistent with the Nemotron 3 Super rollout, which improved inference throughput without meaningful accuracy loss.

Speed Advantage

On a pre-release DeepInfra endpoint, Nemotron 3 Ultra served over 300 tokens per second. Chinese-origin models in its intelligence range — DeepSeek and Kimi variants — are generally served at 50-100 tokens per second today. gpt-oss-120b reaches similar speeds but at 33 on the intelligence index, 15 points lower.

For inference workloads where throughput matters as much as capability, this is the most viable US open-weights entry since Llama 4 Maverick.

Where the Ceiling Is

Nemotron 3 Ultra does not reach the Chinese open-weights frontier. Kimi K2.6 sits at 54 on the same index, and multiple Chinese-lab models cluster in the 52-54 range. The gap is real: 6 points on the AA Intelligence Index represents meaningful capability difference across reasoning and coding evaluations.

That 6-point gap has also been closing. One year ago the Chinese-US open-weights spread was 13 points. Nemotron 3 Ultra narrows it further, but closing the remaining gap will require either a step change in pretraining compute or a model architecture shift.

Key Numbers

ModelAA Intelligence IndexActive ParamsSpeed
Nemotron 3 Ultra4855B300+ tok/s
Kimi K2.654—50-100 tok/s
Gemma 4 31B3931B—
Nemotron 3 Super36——
gpt-oss-120b33—~300 tok/s

Full benchmarks from Artificial Analysis will follow at general availability. NVFP4 weights are expected alongside the BF16 release.