Updated on Jul 21, 2026
• Added the latest RWKV7-G1h model weights and a new RWKV-VL vision-language model
• Improved the chat experience, reasoning display, reasoning-mode descriptions, and screen rotation
• Improved overall performance and responsiveness, along with model selection and offline translation
• Improved stability when continuing responses and using parallel inference, with better real-time Markdown and code rendering
• Fixed crashes on some devices, generation cache issues, and other known problems
