Lightweight
A stripped-down version of the full local model. Gives you a real taste of the beast without the heavy footprint.
Live sessions
Closed-source text model trained on 8B+ parameters,
patched with Anthropic's Neural Engine. Built to perform.
Web engine stripped for speed. Knowledge cutoff Jan 2025. Precision depends on training sources. AI can make mistakes. Session data vanishes when the tab closes. No persistent memory. No telemetry.
Stripped to the essentials. Fast, focused, and private by design.
A stripped-down version of the full local model. Gives you a real taste of the beast without the heavy footprint.
Pure text interface for now. No multimodal overhead — just clean, fast conversation.
Knowledge base tightly scoped to cut resource use. Answers stay accurate up to the January 2025 cutoff.
Works on older browsers and unstable connections. Performance can vary slightly across engines.
No memory between sessions. No telemetry. Close the tab and everything disappears.
Anthropic's architecture hard-coded at the core — the same engine that powers Claude.
Keeping the model online costs real server money. To keep it fair for everyone, requests are capped at 500 per day.
* Limit is global, not per-user.
Open Grox Web