Llama-3.3-Nemotron-Super-49B-v1.5 · NVIDIA · Model History

Claims and evidence

Released 25 July 2025The developer blog header shows Jul 29, 2025, the date of an update with leaderboard information; the post says it first ran on July 25, the Friday on which the model was released. Primary source[1]Release Date; Model Version [2]introduction (latest version released Friday) and closing line (originally ran July 25)

Status AvailableNVIDIA has no deprecation page for its open models; the weights were still downloadable from NVIDIA's Hugging Face organisation on 2026-10-01. Primary source[1]

Successor of Llama-3.3-Nemotron-Super-49B-v1The model card calls it a significantly upgraded version of Llama-3.3-Nemotron-Super-49B-v1; recorded as successor-of rather than revision-of because the version in the name changed to v1.5. Primary source[1]Model Overview
Derived from (distillation) Llama 3.3 (Meta) Primary source[1]derived from Llama-3.3-70B-Instruct

Change · Reasoning Further post-trained on additional reasoning data; NVIDIA says this improves math, science, coding, function calling, instruction following and chat.Compared with Llama-3.3-Nemotron-Super-49B-v1 Primary source[2]
Change · Tool use NVIDIA trained tool calling with iterative DPO stages, and the model card provides a tool-call parser for serving the model with vLLM.Compared with Llama-3.3-Nemotron-Super-49B-v1 Primary source[1]

Input text Primary source[1]Input
Output text Primary source[1]Output
Feature Reasoning modeReasoning on by default; /no_think in the system prompt switches it off. Primary source[1]Quick Start and Usage Recommendations
Feature Function calling Primary source[1]Running a vLLM Server with Tool-call Support
Open weights YesWeights published on Hugging Face under the NVIDIA Open Model License, with the Llama 3.3 Community License Agreement as additional information. Primary source[1]License/Terms of Use; Release Date
Context window 128K tokensThe card also gives the figure as up to 131,072 tokens. Primary source[1]Model Overview; Input
Parameters c. 49BFigure as given in the official model name; the card text gives no separate count. Primary source[1]model name
Access Open-weights download, APIReleased on Hugging Face and as a preview API on build.nvidia.com; the blog said an NVIDIA NIM microservice would follow soon. Primary source[1]Release Date; Quick Start and Usage Recommendations [2]Get started with Llama Nemotron Super v1.5

None recorded.

None recorded.

AI Radar coverage starts in June 2026; no coverage linked yet.