New Free Privacy Focused Android App Allows a Record Breaking 70 Billion Parameter AI Model to Run Entirely Offline on High-End Android Devices

“Stephen J. Pereira, Lead Developer of the project said: “Our software is a bit like a cassette player for optimised AI models, with the different available LLMs considered a collection of cassettes. Android makes that idea even more powerful because the device landscape is so broad.””
The privacy-focused AI app LiberaGPT has officially launched on Android, bringing powerful, fully offline AI to standard Android devices. Alongside its core features, the release introduces a groundbreaking capability that allows compatible high-memory handsets with 24GB ram to run a massive 70-billion parameter large language model entirely on-device, setting a new standard for local mobile computing.

Bedfordshire, UK – June 19th, 2026 – Independent British software house 5N6 today announced the launch of LiberaGPT for Android, marking a new record breaking milestone in mobile artificial intelligence by making it possible to run a 70 billion parameter large language model entirely offline on an unmodified consumer Android handset.

The release represents a major scientific and software engineering breakthrough for private AI. Until recently, models of this size were associated with cloud data centres, specialist desktop hardware, or dedicated AI infrastructure. LiberaGPT now brings that class of model onto compatible Android phones, without modification to the handset, without an external server, and without requiring the user’s prompts or responses to ever leave the device.

LiberaGPT for Android is entirely free to use for local AI chat. There is no subscription, no account requirement, and no paywall on any of the seven large language models in the launch lineup. Every LLM is available to download and run privately on-device.

LiberaGPT also ships ready to use. SmolLM3 3B is bundled inside the app, meaning users have a working private AI assistant immediately after installation, with no model download required before first use.

The launch version includes seven models ranging from 1.5 billion to 70 billion parameters. These include SmolLM3 3B, a built-in dual-mode reasoning model with 128K context; StableLM Zephyr 1.6B, a fast conversational model; AceInstruct 1.5B, NVIDIA’s coding and mathematics specialist with 128K context; EXAONE Deep 2.4B from LG AI, with structured reasoning support; Phi-4 Mini 3.8B from Microsoft, supporting 128K context and function calling; Qwen3 30B MoE, with 30 billion total parameters and 3.3 billion active per token; and Llama 3.3 70B in Record Mode for compatible high memory devices.

This makes LiberaGPT not just another mobile AI app, but a new benchmark for what can be achieved on a private unmodified handset. The 70B Record Mode requires a compatible Android device with sufficient memory, including 24 GB RAM class handsets, while Qwen3 30B MoE requires 16 GB RAM. The rest of the model lineup is designed to make private offline AI available across a much wider range of devices.

LiberaGPT works on most relatively new handsets running Android 12 or later. It is not a flagship-only product. Most phones can run models such as SmolLM3 and StableLM Zephyr on lesser capable hardware while benefiting from the same privacy and offline functionality. More advanced handsets, including devices using Snapdragon 8 Gen 3, Snapdragon 8 Elite, Dimensity 9300, and Dimensity 9400 class chipsets, can unlock GPU acceleration through Vulkan and support larger context windows.

The core achievement is architectural. After model download, LiberaGPT requires zero network connection for inference. No API key, no account, and no internet connection are needed to use the app. It works on a plane, underground, in remote areas, or anywhere signal is unavailable.

Nothing leaves the device. Conversations, prompts, and usage patterns are not transmitted anywhere. There are no analytics SDKs, no telemetry, and no background data collection. The app is designed so that private user input remains private by architecture, not merely by policy.

Context length is another major part of the breakthrough. SmolLM3, Phi-4 Mini, and AceInstruct support context windows up to 131,072 tokens, allowing long document work, extended reasoning, and substantial offline AI sessions directly on a phone.

The app monitors device temperature in real time and adjusts generation behaviour to help protect performance and prevent thermal throttling during longer sessions. This allows more consistent local generation while respecting the limits of mobile hardware.

The comparison with earlier generations of AI highlights the scale of the advance. OpenAI’s GPT-2, introduced in 2019, was widely recognised as a landmark model at 1.5 billion parameters and depended on centralised infrastructure. LiberaGPT for Android now places models from that scale up to 70 billion parameters directly on the handset, privately and offline.

Earlier this month, LiberaGPT for iPhone received a major update introducing unlimited, on-device generation of photorealistic images. Powered by a groundbreaking new model, this feature will also roll out to the Android version next month. The release includes a robust suite of additional tools; such as agentic browsing and a prompt save-and-recall feature, solidifying LiberaGPT’s position as the ultimate ‘Swiss Army knife’ for everyday, privacy-focused offline AI.

Stephen J. Pereira, Lead Developer of the project said: “Our software is a bit like a cassette player for optimised AI models, with the different available LLMs considered a collection of cassettes. Android makes that idea even more powerful because the device landscape is so broad. Some phones are built with huge memory headroom and very capable mobile GPUs, and we can now use that hardware to run models that previously belonged in the cloud. The user’s unique thoughts, ideas and concepts remain their own, rather than becoming another invisible contribution to a centralised training and data infrastructure.”

For this small software firm, the Android launch signals a clear shift in direction. Instead of treating smartphones as thin clients for distant computation, LiberaGPT treats them as capable private computers in their own right. As mobile hardware improves and AI models become more efficient, more everyday intelligence can move from cloud infrastructure back into the user’s pocket.

Beyond the technical achievement, the focus remains practical: private AI that works anywhere. On the train, on a flight, underground, in poor signal areas, or in environments where confidentiality matters, LiberaGPT gives users access to useful AI without needing permission from a network connection or a remote server.

LiberaGPT is available for free on Android through Google Play and on iPhone through the iOS App Store.

Google Play: https://play.google.com/store/apps/details?id=com.liberagpt.app

App Store: https://apps.apple.com/app/liberagpt/id6761083160

Official Website: www.liberagpt.com

Media Contact
Company Name: 5N6 LTD
Contact Person: Amy June Waller
Email: Send Email
Phone: 07860395137
Address:5 Tivoli House Wigmore Park Centre
State: Bedfordshire
Country: United Kingdom
Website: www.5N6.co.uk

 

Press Release Distributed by ABNewswire.com

To view the original version on ABNewswire visit: New Free Privacy Focused Android App Allows a Record Breaking 70 Billion Parameter AI Model to Run Entirely Offline on High-End Android Devices