Ollama runs large language models locally on a Windows PC, so prompts to models such as Llama, Mistral, Qwen or DeepSeek stay on your own machine. Version 0.40.14 was released on November 3, 2026.
Free, MIT licensed, no paid tier listed
Ollama costs nothing and is released under the MIT license.
The installer is 1598.4 MB, a sizeable download before you add any model files.
Local models and the tags that point to them
The publisher describes the app as a way to get large language models running locally. Tags name Llama, Mistral, Qwen and DeepSeek, and the package ID is Ollama.Ollama.
Maintenance in 0.40.14: API and launch fixes
The latest release touches the OpenAI-compatible endpoints, which now keep a model's own sampler defaults instead of forcing 1.0. It also fixes the /v1/responses handling of prior assistant refusals.
The app can now retry ChatGPT disconnect failures, and a faster top-k sampling path was added for MLX.
What's new in 0.40.14
Version 0.40.14 stops OpenAI-compatible endpoints from overriding a model's packaged sampler defaults. It also decodes compressed cloud web search responses on /api/chat correctly.
Ollama pros and cons
Pros
- Free, with an MIT licence
- Download comes from the official github.com host
- Tags cover Llama, Mistral, Qwen and DeepSeek models
- Release notes are detailed and specific
- Updated on the release date, November 3, 2026
Cons
- The installer is large at 1598.4 MB
- Large installer download
- Model files add to the download size
Is Ollama safe?
Ollama is published by the Ollama developer under the MIT license, so the source code can be inspected by anyone. Check the file yourself if that matters to you.
How to install Ollama
- Press the download button to get the official file.
- Run the installer you saved and confirm the Windows security prompt if the publisher matches.
- Follow the setup steps, then start Ollama from the Start menu.
