Ollama v0.40.5 Released: NVIDIA Nemotron 3.5 Support Added

At a Glance
| Item | Value |
|---|---|
| Repository | ollama/ollama |
| Version | v0.40.5 |
| Published | 2026-10-15 |
| License | MIT |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
Ollama, a tool designed to make running large language models locally easy, has released its latest version, v0.40.5. The most significant change in this update is the addition of support for NVIDIA’s new model, “Nemotron 3.5 Super." This allows users to leverage an advanced vision-language model excelling in coding, reasoning, tool use, and document understanding through Ollama.
Breaking Changes & Deprecations
No setting changes that break existing usage or the introduction of deprecated options have been reported in this release.
Key Changes
Support for NVIDIA Nemotron 3.5 Super
The new vision-language model developed by NVIDIA, “Nemotron 3.5 Super," is now available. This model is specialized for coding, reasoning, tool use, and document understanding. It combines a 120B parameter hybrid Mamba-2/Transformer Mixture-of-Experts (MoE) backbone (with 12B active parameters per token) and a C-RADIO vision encoder. It also supports tool calling, an on/off toggleable thinking mode, and contexts of up to 1 million tokens. Users can run this model by executing the following command:
ollama run nemotron-3.5-super
Additionally, the internally used llama.cpp and MLX have been updated accordingly.
Cloud-Based Decision Models
Decision models can now be used in the cloud. Specifically, by using clef:cloud and clef-flash:cloud, it is possible to run models via the ollama.com cloud without pulling them locally. This serves as an effective option when users want to save local storage space or reduce model download times.
Bug Fix in OpenAI-Compatible Conversations
An issue where errors occurred when continuing OpenAI-compatible Responses conversations involving custom tool calls has been fixed. As a result, users continuously utilizing complex conversational flows that involve tool use can expect more stable operation.
Improved Error Messages for Cloud Access Failures
Error messages when cloud access fails have been improved to be easier to understand. They now specifically explain the cause of failure regarding sign-in issues in the desktop app, payment-related issues, and upstream access errors from the ChatGPT gateway. This makes troubleshooting easier for users utilizing cloud features.
Supported Models and Hardware
Nemotron 3.5 Bolt
As a compact sibling model to “Nemotron 3.5 Super," “Nemotron 3.5 Bolt" is now newly supported. This model is a distillation of the “Super" capabilities down to 74B parameters (with 9B active parameters).
“Nemotron 3.5 Bolt" is tuned to fit into a 24 GB NVIDIA GPU, making it suitable for running everyday agent tasks in a local environment. This enables the efficient processing of advanced tasks even in common GPU environments. Furthermore, the model comes bundled with a drafter and supports speculative decoding using it, which aims to improve inference efficiency.
How to Update
Specific commands for updating the Ollama software itself are not specified in the documentation. For updating to the latest version, please check the official release page below:
ollama/ollama v0.40.5 (GitHub)
Related Articles
- Ollama v0.40.15 Released with Parallel Vision GGUF Support
- Ollama v0.40.12 Released: Adds Liquid AI LFM3 and d1 Models
- Ollama v0.40.9 Released with Apple Silicon Phonon Speech Support
- Ollama v0.40.4 Released with Decision Models Support
What to Read Next
- Follow this tool → Ollama overview and release history (200 releases tracked)
- Other inference engines and runtimes → llama.cpp / vLLM / SGLang

