AI Product Updates: New Features and Releases

Sleep app Sleepagotchi is now Gotchi Labs with its token renamed to $CHI, expanding from a single sleep agent to dedicated agents for shopping, fitness, productivity and personal finance.

OpenWorker is a desktop AI coworker that runs locally and produces finished deliverables such as security reviews and incident reports instead of just chat.

Xiaomi released and open-sourced the omni-modal MiMo-V2.6-Pro and Flash models, saying Pro scores 46 on the Intelligence Index, alongside 7,000+ RL task environments and a training framework.

Apps and agents can now run as a Teams bot on Vercel Connect, replying with your code when mentioned in channels or messaged directly.

Grok 4.7 is out and can be used via the Grok Build terminal, the 𝕏 AI website, and Cursor; the author says it works well for making games.

ChatGPT Finances can identify duplicate or overlooked charges and contact customer service on your behalf, with the product lead sharing a case where a hotel stay booked with points was still billed.

Incubated inside Deel, Akai lets you record your screen while narrating a task and converts the demo into an automated workflow the team can review and reuse, already used for tasks like month-end reconciliation.

Xiaomi released the weights and technical report for MiMo V2.6 Pro (1.02T MoE) and Flash, and plans to open about 7,000 Agentic RL training tasks, environments, and a 9B distilled model.

Xiaomi open-sourced the MiMo-V2.6 series, with Pro scoring 46.32 on the Artificial Analysis Intelligence Index to rank first among open-weight models, at the same API price as V2.5.

Pexo is positioned as an AI video agent that generates videos from a user's idea and reference material, iterating the way one would vibe-code.

Synara released a Computer Use beta on macOS and added project import from Codex/Claude, a model picker, native PR management and other features.

Underdog introduced Husky, a model-specific inference engine it says is up to 4.5x faster than Apple's MLX and focused on Pareto-frontier performance.