LFM3 model update • New: Liquid AI's latest LFM3 family — LFM3 1.7B and LFM3 VL 2B, which understands images • LFM3 supports thinking mode, which you can turn on or off anytime • LFM3 can use tools in Agent mode • iPhone Duo support; iPhone Duo and iPad can also run larger models such as Qwen3.5 4B

Support for new models; added Agent model sessions

Two new on-device models: LFM2.5 2.6B for stronger reasoning and tool calling, and LFM2.5 450M VL, a compact vision model that understands images entirely offline. New • LFM2.5 2.6B — available as MLX 4-bit or GGUF Q4_K_M, with reasoning and tool calling built in. • LFM2.5 450M VL — a vision-language model in a 468 MB download. Ask about photos, screenshots and documents without anything leaving your device. • Both models download from Hugging Face or ModelScope. Improved • Smoother model downloading, switching and chat interactions throughout the app. • Fixed known issues from the previous release for a more stable experience.

Added support for the MiniCPM5 1B 4-bit model, including downloads from Hugging Face and ModelScope.

Added Gemma4 model

add LFM2.5-350M , LFM2.5-350M is a 350M-parameter hybrid model for on-device deployment, offering strong performance in a small size with very fast inference.

- add new model

Added VLM (Vision-Language Model) inference support Optimized deep reasoning interaction Introduced Qwen3.5 0.8B Added support for LFM2.5 1.2B and 尝1.6B models, including VLM and Instruct variants UI refactoring with enhanced iOS user experience

- fix bugs

Add a new model Add parameter settings Fix known issues

Resolved the issue of model download failures and optimized offline inference.

- Fix some issues and feature improvements to optimize the experience

- Add Markdown parsing - Modify some problems to optimize the experience

- fix iap issue

support more model

- Added two new subscription plans - Fixed known issues

-Free version OCR recognition problem fixed. -Fixed the issue of ignored instruction templates in the free version.

It is not possible to use OpenAI without configuring an OpenAI API key.

The pro version has added a traffic forwarding service, which can be used simply by configuring a secret key.

TTS supports all languages iPadOS and macOS do not display keyboard close button macOS message background color adjustment

Custom API domain Add various theme colors

- Add chat mode - Add system prompt template - Add model customization - Fix known issues

- Scan translated text - Support Apple Watch