Features
Fill-in-the-blank questions
Open-ended answer evaluation with AI feedback
Spaced repetition scheduling
Audio review (text-to-speech)
Built-in study timer with optional app blocking
WhatsApp study companion (Text Lexie)
Image occlusion for diagrams
Language learning support
PDF and text input support
Offline processing (photos stay on device)
198B-A11B mixture-of-experts architecture
Vision-language understanding and generation
Fast inference for agent tool-use tasks
Multi-turn function calling support
Large context window for agent loops
Open-source model weights on HuggingFace
Strong multimodal reasoning and visual QA
High throughput per GPU with sparse activation
Customizable system prompts for agent behavior
Low-latency callback API design
Step-2 1.5B-step reasoning model on GitHub
