Nemotron 3 Super
A text-only model in NVIDIA's Nemotron 3 series, released in March 2025, and the predecessor generation to Nemotron 3.5 Super VL.
Timeline (last 90 days)
The speaker's personal AI assistant 'Magic' runs on a headless DGX Spark, using the Nemotron 3 Super model (120 billion parameters, only 12 billion active).
On the PinchBench benchmark, Nemotron 3 Super closely trails the major closed-source frontier models.
The author tested Nemotron 3 Super returning in about 3 seconds on NVIDIA's hosted API.
Awas uses Nemotron 3 Ultra on Nebius Token Factory to extract structured fields from ads and uses Nemotron 3 Super to propose evidence rows.
Nemotron 3 Super and Lightning reject the json_schema example in the structured output guide, returning HTTP 422; the name/schema/strict form in the API reference works.
Indonesian-language field accuracy measurements: Nemotron 3 Ultra 84.8%, Nemotron 3 Super 70.7%, Nemotron 3 Lightning 59.6%, with reference labels drafted by an agent and not human-reviewed.
All Facts (6)
The speaker's personal AI assistant 'Magic' runs on a headless DGX Spark, using the Nemotron 3 Super model (120 billion parameters, only 12 billion active).
50%UnverifiedOn the PinchBench benchmark, Nemotron 3 Super closely trails the major closed-source frontier models.
50%UnverifiedThe author tested Nemotron 3 Super returning in about 3 seconds on NVIDIA's hosted API.
50%UnverifiedAwas uses Nemotron 3 Ultra on Nebius Token Factory to extract structured fields from ads and uses Nemotron 3 Super to propose evidence rows.
50%UnverifiedNemotron 3 Super and Lightning reject the json_schema example in the structured output guide, returning HTTP 422; the name/schema/strict form in the API reference works.
50%UnverifiedIndonesian-language field accuracy measurements: Nemotron 3 Ultra 84.8%, Nemotron 3 Super 70.7%, Nemotron 3 Lightning 59.6%, with reference labels drafted by an agent and not human-reviewed.
50%