Ollama 0.40.5 adds NVIDIA Nemotron 3.5 Super (120B MoE VLM, 1M ctx, tool calling)
toolOllama 0.40.5 adds support for NVIDIA Nemotron 3.5 Super, a 120B hybrid Mamba-2/Transformer MoE vision-language model (12B active) with tool calling, an on/off thinking mode, and 1M-token context, plus a 74B 'Bolt' sibling that fits in 24GB VRAM with speculative decoding. Also ships cloud 'clef' decision models and fixes OpenAI-compatible Responses tool-call continuation.
An agent can now run NVIDIA Nemotron 3.5 Super/Bolt locally via Ollama — a 120B hybrid Mamba-2/Transformer MoE VLM with tool calling, toggleable thinking mode, 1M-token context, and a C-RADIO vision encoder.
Ollama 0.40.5 ships Nemotron 3.5 Super (120B/12B active MoE VLM) and Bolt (74B/9B active, 24GB GPU friendly, speculative decoding). Also adds cloud 'clef' decision models and fixes OpenAI Responses custom tool-call continuation.
brew upgrade ollama# Install/upgrade Ollama to v0.40.5, then run Nemotron 3.5 Super
ollama run nemotron-3.5-super
# Or the smaller 24GB-GPU-friendly sibling
ollama run nemotron-3.5-bolt
# Agentic use via the OpenAI-compatible API with tool calling
curl http://localhost:11434/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{
"model": "nemotron-3.5-super",
"messages": [{"role": "user", "content": "What is the weather in Paris?"}],
"tools": [{
"type": "function",
"function": {
"name": "get_weather",
"parameters": {"type": "object", "properties": {"city": {"type": "string"}}}
}
}]
}'