AI provider

Liquid AI

Liquid AI is a private foundation-model company founded in 2023 by researchers from MIT CSAIL. Its current LFM models target local deployment across text, vision, audio and combined inputs.

Models & tools

  • A hosted Ultra decision model that accepts text and images through Liquid's decisions API.

    Launch date not confirmedAPI
  • A hosted decision model that accepts text and images through Liquid's decisions API.

    Launch date not confirmedAPI
  • A text model with agent-focused post-training and a 131,072-token context.

    Launch date not confirmedDeprecated API; check access detailsDownloadable
  • A 24B reasoning model with 2B active parameters.

    Launch date not confirmedAPI · Downloadable
  • A multilingual retrieval model that creates one vector per text.

    Launch date not confirmedDownloadable
  • A 1.92B decision model for text, images and video.

    Launch date not confirmedDownloadable
  • An earlier 3.1B decision model based on LFM2.5-VL-3B.

    Launch date not confirmedDeprecated API; check access detailsDownloadable
  • A 2B-class decision model that adds audio input to Liquid's decision line.

    Launch date not confirmedDownloadable
  • A 587M experimental decision model.

    Launch date not confirmedResearch releaseDownloadable
  • A model that summarizes meeting transcripts on local hardware.

    Launch date not confirmedDownloadable
  • A 350M model that translates between Japanese and English.

    Launch date not confirmedDownloadable
  • LFM2-350M-Math

    Liquid AI · Liquid Nanos

    A small English mathematical reasoning model.

    Launch date not confirmedDownloadable
  • A model that extracts personal information from Japanese text as JSON.

    Launch date not confirmedDownloadable
  • A compact 1.2B general-purpose instruction model.

    Launch date not confirmedDownloadable
  • A June 2026 update to Liquid's specialized 1.2B Japanese chat model.

    Launch date not confirmedDownloadable
  • The reasoning variant of Liquid's 1.2B text model.

    Launch date not confirmedDownloadable
  • Liquid's 230M-parameter text model.

    Launch date not confirmedDownloadable
  • A 350M-parameter text model for small local deployments.

    Launch date not confirmedDownloadable
  • A reasoning-tuned text model with 8.3B total and 1.5B active parameters.

    Launch date not confirmedDownloadable
  • An English speech-and-text model with downloadable weights retained after deprecation.

    Launch date not confirmedDeprecated API; check access detailsDownloadable
  • A 1.5B speech model for Japanese speech-to-speech conversation.

    Launch date not confirmedDownloadable
  • A late-interaction retrieval model that represents each text token separately.

    Launch date not confirmedDownloadable
  • A 230M multilingual masked language encoder for task-specific fine-tuning.

    Launch date not confirmedDownloadable
  • A 350M multilingual masked language encoder for task-specific fine-tuning.

    Launch date not confirmedDownloadable
  • A compact text-and-image model built on the LFM2.5-1.2B base.

    Launch date not confirmedDownloadable
  • A 1.6B-class image extraction model that returns user-defined fields as JSON.

    Launch date not confirmedDownloadable
  • A vision-language model that combines the LFM2.5-2.6B backbone with a SigLIP2 NaFlex encoder.

    Launch date not confirmedDownloadable
  • A compact vision model that pairs LFM2.5-350M with an 86M SigLIP2 NaFlex encoder.

    Launch date not confirmedDownloadable
  • A 450M-class image extraction model that returns user-defined fields as JSON.

    Launch date not confirmedDownloadable
  • A 1.73B text model for local agents and task-specific adaptation.

    Launch date not confirmedAPI · Downloadable
  • A speech-and-text model that accepts text or audio and generates text, speech or both.

    Launch date not confirmedAPI · Downloadable
  • A multimodal model that combines text, image, video and audio input with text and speech output.

    Launch date not confirmedAPI · Downloadable
  • A 1.92B vision-language model built on the LFM3 text backbone.

    Launch date not confirmedAPI · Downloadable