Ollama 0.40.13 brings Liquid AI's LFM3 models to Linux and Windows

Ollama 0.40.13 lets Liquid AI's LFM3 models, including text-to-speech in lfm3-audio and lfm3-omni, run on Linux and Windows as well as Apple silicon. That closes the caveat in the previous release, which supported them on Apple hardware only. The release page dates it 28 October.

What changed

The release is small: its notes hold a single item, and it is the one above. The models are the ones Ollama 0.40.12 introduced. Therelfm3is a 1.7B text model with a 256K-token context window ,lfm3-audiodoes speech recognition, translation and text-to-speech in 18 languages , andlfm3-omnitakes text, images and audio in one model. What 0.40.13 changes is where they run: Linux and Windows machines are now included, not only Apple silicon.

The release was made from commit 04088e7, and its changelog line runs from v0.40.12 to v0.40.13 , so nothing else is listed as new sinceOllama 0.40.12.

What it was before

In 0.40.12 the notes were plain: lfm3 and d1 2b were supported on Apple hardware in that release, and support for Linux and Windows was coming soon. That same release said Liquid AI's d1 models now come in a 2B size, built on LFM3. The 0.40.13 notes speak of the LFM3 models and do not name d1, so do not assume the d1 2B model has moved to Linux or Windows until a release says so.

What it means for you

If you use Windows or Linux, update and try the model. The library page gives the command ollama run lfm3. Before you pull it, check your free memory. The library page lists the model at 1.1GB - 1.7GB to download, with a 256K context window , and says that at 4-bit precision it takes about 1 GB of memory, so it fits on any phone or laptop. Memory is not only the model: the 256K tokens of context count your input and the model's output together , so a very long conversation needs more room than the 1 GB figure. Treat that as a reason to start with short prompts and watch your memory use, not as a number the page gives.

Use it for what it was made for.The page says it is designed for local agents, tool calling, structured outputs and domain-specific fine-tuning , and it lists Arabic among its languages. If you use Hermes Agent, the page givesollama launch hermes --model lfm3. Do not make it your coding agent: the page does not recommend it for agentic coding or complex multi-step reasoning, because of its small size , and for knowledge-heavy tasks it advises retrieval or search.

If you want speech, lfm3-audio and lfm3-omniare named in the notes as running on Linux and Windows now. The notes mention text-to-speech by name, so that is the new thing to try on those systems.

If you are on an Apple silicon Mac,nothing in this release changes your setup; the models ran there already.

How to update. Install 0.40.13 from the release page for your system, then open a terminal and type ollama -v: it prints the version you have installed, which should now be 0.40.13. If it shows an older number, quit Ollama completely, start it again and run the command once more.

The source

The release page on GitHub, Ollama v0.40.13; the earlier Ollama v0.40.12; and the library page for lfm3.