AI Product Updates: New Features and Releases

A new MCP provisions real cloud iPhones and Androids for AI agents, letting them operate a phone like a person after a single prompt.

The index now spans 290 tasks with FrontierCode v1.1 replacing DeepSWE, and Claude Code with Claude Opus 5.5 tops the ranking at 61.7.

Agents can borrow dollars against prepaid inference credits to cover gas, data and API costs instead of selling compute.

Proof of Agent, an AI network run by agents, has reopened mining; users can tell a bot or agent to visit minepoa.com and start mining.

A Thai morning digest covering seven updates including Meet live translation in 63 languages, M365 Copilot Projects, and the OpenAI Decisions API beta.

AWS Continuum for code vulnerabilities is GA, covering the full find-to-fix loop and scoring higher than any other AI agent on CyberGym-E2E.

Kylon's new Annotations feature breaks a chosen video into components and rebuilds it around the user's brand, callable via @kylon_ai on X with component-level team comments.

Browser-use open-sourced Open Duet, where agents learn what matters to you, find useful work and prepare it for approval, runnable with Claude Code or Codex.

NVIDIA shipped Flare, a local inference engine for DGX Spark; a head-to-head test shows 47.6 tok/s decode versus TensorFold's 75.7, 1.5x faster first token at 32k, and temperature locked at 1.0.

XE's testnet is live, letting AI agents hold their own wallets, send zero-fee transfers in seconds, and message other accounts peer-to-peer, all traceable on an explorer.

The new Messages API returns 400 if any sampling parameter is sent, using default sampling instead; message validation is stricter, and the 2023-06-01 version still works.

Chatbase partners with Persona on KYA (Know Your Agent), letting personal agents authenticate with businesses, which set exactly what those agents may do.