AI news
Agents
Anthropic opens private beta of self-hosted Claude Managed Agents
Firms can run the agent harness, sandboxes and session history in their own cloud or data centre, and route model calls via Bedrock, Google or Foundry. Follows the 11 Oct Managed Agents dynamic workflows update.
Communication
Anthropic starts rolling out Claude Meetings beta for Team and Enterprise plans
Claude joins Google Meet calls as a visible participant, takes notes and builds docs and whiteboards while people talk. Teams need to settle consent and outside-guest rules before switching it on.
Communication
DeepSeek opens its app's search, file, speech-to-text and text-to-speech tools via API
Developers can rebuild a full chatbot with voice in and out on deepseek-flash without wiring separate services. Text-to-speech covers 10 languages including English, Arabic and Japanese, at a list price of $0.45 per 10,000 characters.
Agents
Expo launches cloud iOS and Android simulators for coding agents
Agents such as Claude Code and Cursor can test mobile app changes on cloud devices and return recordings as proof, at $0.10 a minute. Reviewers watch the change run instead of rebuilding it themselves.
Models
Google releases Gemini 4 Hydrogen, a fast model for coding and agent work
Priced at $1.50 per million input tokens and $7.50 output across a 1 million-token window, it is rolling out in the Gemini app, Search and the API. A cheaper execution model makes multi-agent workflows easier to cost.
Tools
llama.cpp and ggml reach version 1.0.0 with stable, versioned APIs
Products built on local, private AI now get a promise that breaking changes arrive only with a major version bump. The release also adds an OpenAI-compatible speech endpoint and FP8 caches.
Agents
OpenAI launches ChatGPT Wallet so its dots agents can pay for purchases
Agents pay with network tokens rather than card details, and every purchase needs a biometric check. Retailers that block signed agent traffic will miss these sales; it is US Pro users only for now.
Language
Riverside Labs launches Tonality, a voice model that edits single words in recordings
Producers can fix a name, swap a sponsor read or stretch a line for dubbing without re-recording, keeping the speaker's voice and room sound. API access is through sales, and the precision raises consent questions for anyone editing other people's speech.