What’s New in Ollama 0.40.12: Liquid AI’s LFM3 and d1 Models

Ollama 0.40.12 adds Liquid AI’s new LFM3 family, small models built for local agents, tool calling and structured output on a “tri-hybrid” architecture.

The models

• lfm3: a 1.7B text model with a 256K-token context window.
• lfm3-vl adds images, lfm3-audio handles speech recognition, translation and TTS in 18 languages, and lfm3-omni does text, images and audio.
• d1 decision models now come in a 2B size built on LFM3, with d1-omni adding audio.

Heads-up: lfm3 and d1 2b are supported only on Apple hardware in this release. Linux and Windows support is “coming soon.”

Other changes

• Faster prompt processing on Apple silicon for Gemma 4, Flint and Laguna.
• DFlash2 speculative-decoding drafts on MLX, and ollama launch now works with OpenCode v2.
• Updated MLX and llama.cpp.

Review before you run anything: read the docs yourself, back up your config, and try it somewhere you can afford to break first.

God Bless, and Happy hacking!

References & Further Reading

• Ollama v0.40.12 release notes (GitHub)