NVIDIA released Nemotron 3.5 Super as an early-access model. It handles text, image and video inputs. It supports agentic, coding, tool-calling and multimodal tasks. The model offers 262,000 token context. It is available via free and partner endpoints. Download is not available. It supports parallel tool calls. It can stream reasoning and final content separately.
What it unlocks
Running agentic, coding, tool-calling and multimodal workloads with a single model.
- 262,000 context length
- 124B parameters
- 16,384 max tokens
What you need to act on it
- early access
Sources