53 open-source models, all of them on the phone.

Latest Qwen with hybrid architecture for detailed answers and multilingual conversations in 201 languages.

Efficient hybrid model for everyday tasks with wide language support (201 languages).

Compact hybrid model for quick tasks in 201 languages under 600MB.

Analyzes images, reads text from screenshots, and answers questions about photos in 32 languages.

Handles complex questions, multi-step instructions, and conversations in 29+ languages.

Advanced image analysis for charts, diagrams, documents, and complex visual reasoning tasks.

Writes, explains, and refactors code in 92+ programming languages.

Shows step-by-step reasoning for math, logic puzzles, and complex problems.

Handles translation, creative writing, and longer documents in 29+ languages.

Balanced model for everyday tasks and multilingual conversations.

Fast code explanations, bug fixes, and completions in 92+ programming languages.

Everyday conversations, writing help, and basic tasks in 29+ languages.

Handles quick questions, simple conversations, and basic tasks.

Ultra-fast responses for very simple questions when speed matters most.

Strong reasoning and detailed explanations for complex conversations in 8 languages.

Fast and lightweight for summarization, Q&A, and basic tasks in 8 languages.

Excels at math, logic, and reasoning - competes with much larger models on benchmarks.

Solid math and logic performance trained on high-quality textbook data.

Google's mobile-first multimodal Gemma. Understands images and thinks through hard questions, with QAT weights that hold quality at this size.

Multimodal Gemma 4 at full quality. Understands images and thinks through complex questions before answering.

Reasoning and chat model with efficient memory usage for general conversations.

Top-tier reasoning and chat quality with QAT-optimized outputs at same size. Great for general conversations.

Supports 140+ languages in under 1GB - great for less common languages.

Reliable general chat with precise instruction following.

Specialized for code completion and fill-in-the-middle tasks.

Extremely small (253MB) for very basic tasks when storage is critical.

Shows detailed step-by-step reasoning before answering. Great for learning.

Fast code help in 87 languages. Lightweight but has older training data.

Writes and explains code in 18 languages with good context for code review.

Reliable general chat with consistent, helpful responses.

Detailed answers and strong instruction following in multiple languages.

Fast code completions and debugging in 52 languages under 1GB.

Strong chat performance for longer conversations.

Handles complex multi-step tasks with strong instruction following and low memory usage.

Reliable performance for business and professional tasks under 1GB.

Basic tasks with multilingual support at only 223MB.

Punches above its weight with strong answers and instruction following under 2GB.

Good conversations and instruction following at just over 1GB.

Quick responses and basic reasoning at only 390MB.

Tiny model for quick testing. Download a featured model for better responses.

Historic GPT-2 from 2019 - the model that started the modern AI wave. Text completion only (not chat). Interesting for educational purposes and AI history.

Analyzes images and answers directly, with the fast, memory-light inference Liquid models are built for.

Liquid's flagship text model: strong instruction following and multilingual chat with fast, memory-light inference.

Fast and smart in a tiny package. Handles everyday questions and tasks with surprisingly good quality.

Works through problems step by step before answering, in a package small enough for any phone.

PrismML's largest ternary Bonsai model. 8B-scale answers from 1.58-bit weights engineered for fast on-device inference.

PrismML's flagship 1-bit Bonsai model. An 8B-scale download smaller than most 1B models, built for on-device speed.

Experimental 4B-class model compressed with ternary weights into a phone-friendly download.

Experimental compact ternary model for fast local chat on lower-memory phones.

Roleplay and creative writing with consistent character voices.

Balanced 4B model with fast responses for general tasks.

Basic chat under 700MB - runs fast on almost any device.

Ultra-fast responses for basic tasks with multilingual support.