Ollama recently released version 0.40.5, introducing support for NVIDIA's Nemotron 3.5 Super and Nemotron 3.5 Bolt vision-language models (VLMs). Nemotron 3.5 Super is a 120-billion-parameter hybrid Mamba-2/Transformer mixture-of-experts (MoE) model, utilizing a C-RADIO vision encoder. It is designed for coding, reasoning, tool use, and document understanding, supporting tool calling, an on/off thinking mode, and contexts up to 1 million tokens. Its compact sibling, Nemotron 3.5 Bolt, is a 74-billion-parameter model with 9 billion active parameters. It is tuned to run efficiently on a 24 GB NVIDIA GPU for everyday agentic tasks and incorporates speculative decoding via its included drafter. This update expands Ollama's local LLM ecosystem with new, capable VLMs. The release also notes that decision models are now available on cloud via "clef:cloud".