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

Published Oct 24, 2026

01Common complaints

1. You are an enterprise that requires SOC 2/ISO evidence during procurement - Liquid’s public site did not clearly publish these certifications at the time of review. 2. You need a fully permissive open-source license for commercial use at any scale - Liquid’s LFM Open License includes a $10M annual revenue threshold for commercial use under the free terms. 3. Your use case is heavily knowledge- or coding-intensive on-device - LFM2 model documentation explicitly states it is not recommended for knowledge-intensive tasks or those requiring programming skills. 4. The organization already has a stable local inference stack and only needs a runtime - If an organization only requires a native runner for local inference, runtimes such as Ollama, llama.cpp, or MLX may be simpler operationally than adopting a broader toolchain. Sources: LFM license; Hugging Face LFM2 model card; Ollama pricing.

Disqualifiers (poor fit)

Primary need is frontier reasoning or coding tasks on device, where small models typically underperform larger cloud models.
Requirement for publicly documented SOC 2/ISO security posture as part of procurement.
Requirement for fully permissive licensing (no revenue threshold).

Vendor-acknowledged limitations

LFM2 model documentation indicates it is not recommended for knowledge-intensive tasks or those requiring programming skills.

Source: Hugging Face LFM2-2.6B model card.

Product maturity limitations

LEAP was described in launch materials as an early-stage release, indicating potential for API churn and missing enterprise features.

Source: LEAP launch press release.

Usability issues reported in community posts

Community reports include input handling complaints for Apollo (e.g., missing send button behavior on Android builds).

Source: Reddit thread.

Licensing limitation (strategic constraint)

Free commercial rights under the LFM Open License end above $10M annual revenue and require a commercial license thereafter.

Source: LFM Open License.

02Why they buy

Unique selling points supported by available evidence

- An edge-first integrated platform (LEAP) that combines model selection, bundling, SDK deployment, and on-device testing via Apollo. - A model architecture described in technical documentation as combining convolutions and grouped query attention to improve efficiency on device classes. - Support for long context windows (documented 32,768 tokens for certain LFM2 sizes), which is relatively large for small, on-device models. Sources: LEAP documentation; Hugging Face LFM2 documentation and model card.

Areas where competitors may have the advantage

Licensing simplicity: Apache 2.0 models from some competitors avoid revenue thresholds.
Ecosystem integration and runtime maturity: other local-inference projects have broader integration paths and community usage.

Sources: Mistral licensing documentation; Ollama pricing.

03Field reports

Review-platform presence (G2, Capterra, TrustRadius)

No meaningful G2/Capterra/TrustRadius product review pages were found for Liquid AI as a B2B SaaS product (likely due to the company being early-stage and the offering comprising both open-weight models and developer tooling).

As a result, review evidence is drawn from App Store reviews (Apollo) and community posts.

Apollo iOS App Store reviews (subject to app-store sampling bias)

Apollo – Powered by Liquid: very highly rated on the App Store. Full review text was not exhaustively crawled for this report; the aggregate rating indicates generally positive sentiment among app reviewers.

Important caveat: App Store ratings often reflect UI/UX expectations and may not represent enterprise readiness.

Community and social feedback (includes negative items)

A Reddit thread about Apollo Android includes both announcements and user criticisms. Examples: - Negative UX complaint (user quote): "There is no Send button. Enter on keyboard inserts new line. Zero stars." - Alternative preference (user quote): "I think I’ll stick with Gemma 3n-E4B-IT from hugging face for on device AI." Notes: This is a small sample and subreddit context matters; community comments are a limited signal for enterprise evaluation.

Reasons customers may choose Liquid AI (based on vendor materials and ecosystem signals)

On-device latency and privacy objectives.
Model efficiency and ability to run within CPU/NPU constraints.
For some developers, LEAP is presented as reducing the complexity of on-device deployment and model packaging.

Reasons customers might not choose or might churn (observed or likely)

Observed issues: - Apollo UI/UX complaints in community posts (e.g., input handling behavior). Likely churn drivers (inferred): - Organizations standardized on existing local-inference stacks (Ollama and llama.cpp) may view LEAP as an additional toolchain and choose not to adopt it. - The licensing revenue threshold may lead some organizations to select fully permissive alternatives when scaling. - Use cases requiring cloud-scale reasoning, advanced coding assistance, or other tasks that small edge models do not handle well may prefer larger cloud models.

E‑commerce partnership (vendor-announced; details limited)

The vendor announced a multi-year partnership with a major e-commerce platform that reportedly includes: - licensing and deployment of Liquid’s models in search workflows, - a vendor-announced low-latency text model used for search, - references to co-developed recommender work and controlled testing with reported conversion improvements. Critical evaluation: - The announcement is vendor-published and does not include full methodology, dataset details, or independent validation. - One partnership does not guarantee repeatability across different customers or domains. Sources: vendor press release and blog post about the partnership.

LFM2 open release (technical reproducibility signal, but vendor-led)

Liquid AI released LFM2 and published a technical report. Open-weight availability on model hubs allows external inspection.

Critical evaluation:

Open weights increase inspectability and reproducibility potential.
Performance claims depend on implementation details, hardware, and tasks; independent benchmarking remains important.

04Best fit

Best-fit company sizes (practical estimate)

Startups to mid-market companies building privacy-sensitive mobile or edge experiences, particularly those under the vendor's commercial revenue threshold for simplified use of LFM models.
Enterprise buyers with strong latency and privacy requirements may also fit but are likely to require formal security/compliance documentation and a commercial agreement.

Industries / verticals with strongest resonance

E-commerce (vendor partnership signals relevance).
Consumer apps that require offline assistants or on-device features.
Automotive, robotics, and IoT (edge constraints; and the company's robotics research background is relevant).

Roles and personas

Decision makers: Head of Product, VP Engineering, CTO, AI/ML leadership.
End users: mobile engineers, ML engineers, applied scientists, MLOps and edge inference engineers.

Qualification criteria (strong fit)

Need for offline or low-latency inference.
Data privacy constraints that limit sending inputs to cloud services.
Hardware constraints (CPU/NPU, memory-limited devices).
Willingness to pilot or adopt a new toolchain to reduce edge deployment friction.

05The pitch

What the product does (objective value proposition)

Liquid AI is a foundation-model company focused on low-latency, on-device (edge) AI optimized for resource-constrained hardware. Its offerings span: - Liquid Foundation Models (LFMs): open-weight small foundation models optimized for CPU/GPU/NPU inference and long context. - LEAP (Liquid Edge AI Platform): a developer toolchain/SDK to select, test (informally "vibe-check"), bundle, fine-tune, and deploy models on-device (mobile and laptops). - Apollo: a consumer/developer-facing mobile app that demonstrates and tests local models, and also functions as an LLM client that can connect to external providers or self-hosted backends. Liquid AI’s stated position is that many AI workloads do not require large cloud-hosted transformer models, and that architecture choices plus device-aware optimizations can enable low-latency, privacy-preserving, continuously available on-device AI for enterprise and consumer applications. Key vendor positioning statements (presented as vendor claims, not independently verified): - A vendor-announced low-latency production search model used in an e-commerce workflow. - Vendor claims of faster decode and prefill performance versus other models on CPU for LFM2 (vendor blog/model card). - The vendor has used superlative language in press materials (for example, describing the models as "world’s fastest and best-performing open-source small foundation models"). Sources: vendor homepage, press and blog; Hugging Face LFM2 documentation and model card.

Target market and market positioning

- Target users: mobile application developers, product teams, and enterprises seeking on-device AI (privacy/latency/offline requirements) without building an entire ML toolchain. - Positioning: efficient foundation models for edge combined with a toolchain (LEAP) to operationalize them. This offering overlaps with, but is distinct from, several adjacent categories: - edge ML training platforms (e.g., Edge Impulse) - local inference runtimes (llama.cpp, MLX, Ollama) - frontier model APIs (OpenAI/Anthropic) Sources: vendor site and LEAP documentation; competitor pricing pages.

Get a comprehensive analysis of Liquid AI including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →

Published Oct 24, 2026

01Common complaints

You are an enterprise that requires SOC 2/ISO evidence during procurement
- Liquid’s public site did not clearly publish these certifications at the time of review.
You need a fully permissive open-source license for commercial use at any scale
- Liquid’s LFM Open License includes a $10M annual revenue threshold for commercial use under the free terms.
Your use case is heavily knowledge- or coding-intensive on-device
- LFM2 model documentation explicitly states it is not recommended for knowledge-intensive tasks or those requiring programming skills.
The organization already has a stable local inference stack and only needs a runtime
- If an organization only requires a native runner for local inference, runtimes such as Ollama, llama.cpp, or MLX may be simpler operationally than adopting a broader toolchain.

Sources: LFM license; Hugging Face LFM2 model card; Ollama pricing.

Disqualifiers (poor fit)

Primary need is frontier reasoning or coding tasks on device, where small models typically underperform larger cloud models.
Requirement for publicly documented SOC 2/ISO security posture as part of procurement.
Requirement for fully permissive licensing (no revenue threshold).

Vendor-acknowledged limitations

LFM2 model documentation indicates it is not recommended for knowledge-intensive tasks or those requiring programming skills.

Source: Hugging Face LFM2-2.6B model card.

Product maturity limitations

LEAP was described in launch materials as an early-stage release, indicating potential for API churn and missing enterprise features.

Source: LEAP launch press release.

Usability issues reported in community posts

Community reports include input handling complaints for Apollo (e.g., missing send button behavior on Android builds).

Source: Reddit thread.

Licensing limitation (strategic constraint)

Free commercial rights under the LFM Open License end above $10M annual revenue and require a commercial license thereafter.

Source: LFM Open License.

02Why they buy

Unique selling points supported by available evidence

An edge-first integrated platform (LEAP) that combines model selection, bundling, SDK deployment, and on-device testing via Apollo.
A model architecture described in technical documentation as combining convolutions and grouped query attention to improve efficiency on device classes.
Support for long context windows (documented 32,768 tokens for certain LFM2 sizes), which is relatively large for small, on-device models.

Sources: LEAP documentation; Hugging Face LFM2 documentation and model card.

Areas where competitors may have the advantage

Licensing simplicity: Apache 2.0 models from some competitors avoid revenue thresholds.
Ecosystem integration and runtime maturity: other local-inference projects have broader integration paths and community usage.

Sources: Mistral licensing documentation; Ollama pricing.

03Field reports

Review-platform presence (G2, Capterra, TrustRadius)

As a result, review evidence is drawn from App Store reviews (Apollo) and community posts.

Apollo iOS App Store reviews (subject to app-store sampling bias)

Apollo – Powered by Liquid: very highly rated on the App Store. Full review text was not exhaustively crawled for this report; the aggregate rating indicates generally positive sentiment among app reviewers.

Important caveat: App Store ratings often reflect UI/UX expectations and may not represent enterprise readiness.

Community and social feedback (includes negative items)

A Reddit thread about Apollo Android includes both announcements and user criticisms. Examples:

Negative UX complaint (user quote): "There is no Send button. Enter on keyboard inserts new line. Zero stars."
Alternative preference (user quote): "I think I’ll stick with Gemma 3n-E4B-IT from hugging face for on device AI."

Notes: This is a small sample and subreddit context matters; community comments are a limited signal for enterprise evaluation.

Reasons customers may choose Liquid AI (based on vendor materials and ecosystem signals)

On-device latency and privacy objectives.
Model efficiency and ability to run within CPU/NPU constraints.
For some developers, LEAP is presented as reducing the complexity of on-device deployment and model packaging.

Reasons customers might not choose or might churn (observed or likely)

Observed issues:

Apollo UI/UX complaints in community posts (e.g., input handling behavior).

Likely churn drivers (inferred):

Organizations standardized on existing local-inference stacks (Ollama and llama.cpp) may view LEAP as an additional toolchain and choose not to adopt it.
The licensing revenue threshold may lead some organizations to select fully permissive alternatives when scaling.
Use cases requiring cloud-scale reasoning, advanced coding assistance, or other tasks that small edge models do not handle well may prefer larger cloud models.

E‑commerce partnership (vendor-announced; details limited)

The vendor announced a multi-year partnership with a major e-commerce platform that reportedly includes:

licensing and deployment of Liquid’s models in search workflows,
a vendor-announced low-latency text model used for search,
references to co-developed recommender work and controlled testing with reported conversion improvements.

Critical evaluation:

The announcement is vendor-published and does not include full methodology, dataset details, or independent validation.
One partnership does not guarantee repeatability across different customers or domains.

Sources: vendor press release and blog post about the partnership.

LFM2 open release (technical reproducibility signal, but vendor-led)

Liquid AI released LFM2 and published a technical report. Open-weight availability on model hubs allows external inspection.

Critical evaluation:

Open weights increase inspectability and reproducibility potential.
Performance claims depend on implementation details, hardware, and tasks; independent benchmarking remains important.

04Best fit

Best-fit company sizes (practical estimate)

Startups to mid-market companies building privacy-sensitive mobile or edge experiences, particularly those under the vendor's commercial revenue threshold for simplified use of LFM models.
Enterprise buyers with strong latency and privacy requirements may also fit but are likely to require formal security/compliance documentation and a commercial agreement.

Industries / verticals with strongest resonance

E-commerce (vendor partnership signals relevance).
Consumer apps that require offline assistants or on-device features.
Automotive, robotics, and IoT (edge constraints; and the company's robotics research background is relevant).

Roles and personas

Decision makers: Head of Product, VP Engineering, CTO, AI/ML leadership.
End users: mobile engineers, ML engineers, applied scientists, MLOps and edge inference engineers.

Qualification criteria (strong fit)

Need for offline or low-latency inference.
Data privacy constraints that limit sending inputs to cloud services.
Hardware constraints (CPU/NPU, memory-limited devices).
Willingness to pilot or adopt a new toolchain to reduce edge deployment friction.

05The pitch

What the product does (objective value proposition)

Liquid AI is a foundation-model company focused on low-latency, on-device (edge) AI optimized for resource-constrained hardware. Its offerings span:

Liquid Foundation Models (LFMs): open-weight small foundation models optimized for CPU/GPU/NPU inference and long context.
LEAP (Liquid Edge AI Platform): a developer toolchain/SDK to select, test (informally "vibe-check"), bundle, fine-tune, and deploy models on-device (mobile and laptops).
Apollo: a consumer/developer-facing mobile app that demonstrates and tests local models, and also functions as an LLM client that can connect to external providers or self-hosted backends.

Liquid AI’s stated position is that many AI workloads do not require large cloud-hosted transformer models, and that architecture choices plus device-aware optimizations can enable low-latency, privacy-preserving, continuously available on-device AI for enterprise and consumer applications.

Key vendor positioning statements (presented as vendor claims, not independently verified):

A vendor-announced low-latency production search model used in an e-commerce workflow.
Vendor claims of faster decode and prefill performance versus other models on CPU for LFM2 (vendor blog/model card).
The vendor has used superlative language in press materials (for example, describing the models as "world’s fastest and best-performing open-source small foundation models").

Sources: vendor homepage, press and blog; Hugging Face LFM2 documentation and model card.

Target market and market positioning

Target users: mobile application developers, product teams, and enterprises seeking on-device AI (privacy/latency/offline requirements) without building an entire ML toolchain.
Positioning: efficient foundation models for edge combined with a toolchain (LEAP) to operationalize them. This offering overlaps with, but is distinct from, several adjacent categories:
- edge ML training platforms (e.g., Edge Impulse)
- local inference runtimes (llama.cpp, MLX, Ollama)
- frontier model APIs (OpenAI/Anthropic)

Sources: vendor site and LEAP documentation; competitor pricing pages.

Get a comprehensive analysis of Liquid AI including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →