Liquid AI

Name: Liquid AI
Brand: Liquid AI

AI Native

Liquid AI is a foundation-model company that provides open-weight small foundation models, a developer edge toolchain (LEAP) for selecting, bundling, fine-tuning and deploying models on-device, and a consumer/developer mobile app (Apollo) for testing and running local models with device-aware optimizations and long-context support.

Est 2023|Valuation $316.6M|Raised $296.6M|6 Ref Customers

LinkedIn 21K|Web traffic 58K|FTEs 78

MarketsFoundation Models

01Competitive Landscape

Narrative analysis

Areas where Liquid AI shows strengths: - Integrated edge workflow: LEAP is positioned as an on-device AI toolchain including model search, library, bundling, and mobile SDK integration. - Model architecture targeted at devices: available documentation describes hybrid blocks (short convolutions + attention) aimed at improving efficiency on embedded SoCs and CPUs. Sources: LEAP platform documentation; Hugging Face LFM2 documentation. Where competitors may be stronger: - Licensing clarity and simplicity: Mistral's open models use Apache 2.0 licensing, which may simplify enterprise legal review compared with revenue-threshold-based licenses. - Local-inference ecosystem maturity: for certain hardware (e.g., Apple devices), MLX, llama.cpp, and other runtimes have established integrations and community benchmarking to guide tradeoffs between throughput and time-to-first-token. Sources: Mistral licensing documentation; OSI commentary; academic comparisons of local inference runtimes. Pricing/value comparison insight: - Liquid's revenue-threshold license may be attractive for research groups, nonprofits, and early-stage startups under the threshold. - It may be a disadvantage for scale-ups and enterprises that prefer permissive terms without revenue-based triggers. Source: LFM Open License.

	Liquid AI (LFMs + LEAP)	Ollama	Mistral open models
Starting Price	No standard posted tiers; Apollo app is free; model license free until a commercial revenue threshold, after which a commercial license is required	Free local; optional cloud plans are described in public materials	Model weights published under Apache 2.0 (per Mistral documentation); hosted APIs and pricing are separate and not detailed here
Best For	Teams needing packaged on-device deployment, model selection/bundling, and resource-optimized small models; organizations requiring low-latency custom deployments	Developers requiring a straightforward local model runtime with ecosystem integrations; optional cloud convenience	Teams that require permissively licensed open models usable commercially without a revenue threshold; self-hosting and customization
Key Strength	Device-aware edge focus; an integrated toolchain (LEAP) plus models designed for resource-constrained inference and long context	Simpler local runtime and broad community usage; clear pricing signals for hosted/cloud convenience	Permissive licensing (Apache 2.0) for open models; community adoption and transparency
Key Weakness	License revenue threshold for commercial use above a set limit; limited public enterprise security documentation; early-stage tooling risk	Not specialized as a mobile-first edge SDK out of the box; runtime and model performance vary by environment	Not a dedicated edge deployment platform; small models may still require additional engineering to reach target on-device latency

01Competitive Landscape

Narrative analysis

Areas where Liquid AI shows strengths:

Integrated edge workflow: LEAP is positioned as an on-device AI toolchain including model search, library, bundling, and mobile SDK integration.
Model architecture targeted at devices: available documentation describes hybrid blocks (short convolutions + attention) aimed at improving efficiency on embedded SoCs and CPUs.

Sources: LEAP platform documentation; Hugging Face LFM2 documentation.

Where competitors may be stronger:

Licensing clarity and simplicity: Mistral's open models use Apache 2.0 licensing, which may simplify enterprise legal review compared with revenue-threshold-based licenses.
Local-inference ecosystem maturity: for certain hardware (e.g., Apple devices), MLX, llama.cpp, and other runtimes have established integrations and community benchmarking to guide tradeoffs between throughput and time-to-first-token.

Sources: Mistral licensing documentation; OSI commentary; academic comparisons of local inference runtimes.

Pricing/value comparison insight:

Liquid's revenue-threshold license may be attractive for research groups, nonprofits, and early-stage startups under the threshold.
It may be a disadvantage for scale-ups and enterprises that prefer permissive terms without revenue-based triggers.

Source: LFM Open License.

	Liquid AI (LFMs + LEAP)	Ollama	Mistral open models
Starting Price	No standard posted tiers; Apollo app is free; model license free until a commercial revenue threshold, after which a commercial license is required	Free local; optional cloud plans are described in public materials	Model weights published under Apache 2.0 (per Mistral documentation); hosted APIs and pricing are separate and not detailed here
Best For	Teams needing packaged on-device deployment, model selection/bundling, and resource-optimized small models; organizations requiring low-latency custom deployments	Developers requiring a straightforward local model runtime with ecosystem integrations; optional cloud convenience	Teams that require permissively licensed open models usable commercially without a revenue threshold; self-hosting and customization
Key Strength	Device-aware edge focus; an integrated toolchain (LEAP) plus models designed for resource-constrained inference and long context	Simpler local runtime and broad community usage; clear pricing signals for hosted/cloud convenience	Permissive licensing (Apache 2.0) for open models; community adoption and transparency
Key Weakness	License revenue threshold for commercial use above a set limit; limited public enterprise security documentation; early-stage tooling risk	Not specialized as a mobile-first edge SDK out of the box; runtime and model performance vary by environment	Not a dedicated edge deployment platform; small models may still require additional engineering to reach target on-device latency