← All companies

Frontier / Company profile

Liquid AI

What Liquid AI does

Liquid AI is an efficiency-first foundation-model company spun out of MIT CSAIL (est. 2023) that builds “Liquid Foundation Models (LFMs)” and supporting software to make those models practical across cloud and edge deployments. The company positions its core differentiation as alternative neural architectures and co-design across model + inference so the models can run with lower compute and better latency/cost characteristics than frontier transformer-style systems, including on-device use cases. Liquid sells to developers and enterprises via its model/catalog experiences and an edge deployment platform, and it also engages in partner-led integration work for specific environments (e.g., in-car and Mac on-device assistants). Liquid’s product surface includes Liquid Edge AI Platform (LEAP), Liquid Apollo (a local/low-latency playground), a public model playground, and distribution via Hugging Face; the company also publishes frequent engineering and model-release updates across its LFM2.x model line (e.g., quantization/distillation, speculative decoding, and vision-language variants).

News

Company record

Aug 20, 2026 · Official · Liquid AILFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook

Liquid released DSpark draft model checkpoints for multiple LFM2.5 models and reported throughput/latency improvements from H100-class GPUs to MacBook-class devices, including speculative decoding support for the LFM ecosystem.

Aug 12, 2026 · Official · Liquid AILFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge

Liquid released LFM2.5-VL-3B, describing improved screen/UI understanding, function calling, grounding, and multi-image input performance for edge deployments; it reported benchmark lifts and emphasized low-latency direct answering (non-reasoning).

Aug 05, 2026 · Official · Liquid AIMacPaw Partners with Liquid AI to Bring On-Device AI to Millions of Mac Users

MacPaw and Liquid announced a long-term partnership combining Liquid Foundation Models with MacPaw’s Elix and Mnemos technologies for an on-device Mac assistant; Liquid positioned the integration as enabling private, fast, offline-capable AI experiences on macOS.

Aug 04, 2026 · Official · Liquid AILFM2.5-2.6B: Deploy Agents Everywhere

Liquid released LFM2.5-2.6B, an agentic model intended to run entirely on-device, with claims that it is small enough for phones, responsive on CPUs, and capable for multi-step tool-using workflows without cloud APIs.

Apr 23, 2026 · Official · Liquid AILiquid AI and Mercedes-Benz partner to scale embedded in-car intelligence

Mercedes-Benz announced a multi-year partnership with Liquid AI to scale embedded on-device intelligence for North America, targeting MBUX (third- and fourth-generation) experiences and aiming for a production deployment path in 2026.

Mar 03, 2026 · Official · Liquid AILiquid AI and Insilico Medicine Announce Strategic Partnership Delivering Lightweight Scientific Foundation Models for Drug Discovery

Liquid and Insilico Medicine announced a partnership producing LFM2-2.6B-MMAI, a single lightweight scientific foundation model for drug discovery tasks, designed for local/private infrastructure deployments; Liquid reported training on pharmaceutical data totaling ~120B tokens across 200+ tasks.

Dec 13, 2024 · Official · Liquid AIWe raised $250M to scale capable and efficient general-purpose AI

Liquid announced a $250M Series A to accelerate development, scaling, and deployment of Liquid Foundation Models, including scaling compute infrastructure and improving product readiness for edge and on-prem.

Dec 06, 2023 · Official · Liquid AILiquid AI: A New Generation of AI Models from First Principles

Liquid emerged from stealth and announced its mission and seed financing, describing LFMs built from first principles and led seed capital from OSS Capital and PagsGroup.

Source map · 2 recurring channels · 19 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuation$2B
Tracked capital$296.6M2 sourced rounds

What it builds

Liquid Edge AI Platform (LEAP)

LEAP is Liquid’s platform for discovering, specializing, and deploying Liquid models on-device and in edge/on-prem environments, including workflows for model search, testing/comparison, fine-tuning CLI tooling, and deployment via an edge SDK.

source ↗
Liquid Apollo

Liquid Apollo is Liquid’s low-latency, cloud-free playground intended for interactive local use (“in your pocket”) without a barrier to entry.

source ↗
Liquid Foundation Models (LFMs)

Liquid Foundation Models are Liquid’s general-purpose and task-specific foundation models (including text and vision-language variants in the LFM2.x line) intended to be efficient and suitable for deployment across different compute constraints.

source ↗
Model Playground

Liquid’s model playground is a public interface for trying and exploring Liquid models (linked from the company site and LEAP navigation).

source ↗
Distribution on Hugging Face

Liquid provides model checkpoints for LFM releases through Hugging Face as an access path for downloading, running, and fine-tuning models locally.

source ↗

Key metrics

Milestones & partnerships