Ollama 0.40.2 to 0.40.6: Nemotron 3.5 Super and Bolt, model backup copies

Ollama published a run of releases in October 2026. v0.40.5 (Oct 14) adds NVIDIA Nemotron 3.5 Super (`ollama run nemotron-3.5-super`). The notes describe it as a vision-language model for coding, reasoning, tool use and document understanding. They say it pairs a 120B-parameter hybrid Mamba-2/Transformer mixture-of-experts backbone (12B active per token) with a C-RADIO vision encoder, and supports tool calling, an on/off thinking mode and context up to 1M tokens. The same notes list Nemotron 3.5 Bolt, which distills Super to 74B parameters (9B active). Ollama says Bolt is tuned to fit a 24 GB NVIDIA GPU for everyday agentic tasks, with speculative decoding through its included drafter. v0.40.5 also updates llama.cpp (adding Nemotron 3.5 support) and MLX. It fixes an error when continuing OpenAI-compatible Responses conversations that include custom tool calls. The v0.40.2 notes (Oct 8) say models downloaded with earlier versions of Ollama are upgraded in the background the first time they run, for better performance and compatibility on llama.cpp. To make downgrading safe, Ollama keeps the original copy on disk as a backup. A future release will remove these backups automatically. The notes include a script (requires `jq`) that deletes only the backups. After that, downgrading below 0.40 means re-pulling those models. v0.40.4 (Oct 12) improves Gemma 4 prefill on NVIDIA GPUs with the MLX runner and adds multimodal embeddings through llama-server. It improves speculative decoding reliability in the MLX runner and lets `ollama run` use decision models with `--ask`. v0.40.6 (Oct 16) adds the q2_0 GGUF tensor and file types and prevents orphaned runners on Windows. v0.40.7-rc0 (Oct 18) is a pre-release only.