OpenAI is turning voice into an input for directing multiple desktop tasks while Anthropic adds stronger model choices and app connections; both approaches extend voice interaction from conversation into workflows.
Artificial Intelligence··Morning
Voice connects to tasks on the desktop
OpenAI announced that it added ChatGPT Voice to its Work and Codex experiences on macOS and Windows: users can start tasks, ask about progress and direct work within the selected experience's permissions, with only one conversation running at a time. Anthropic expanded beyond Haiku by adding Opus and Sonnet choices plus Gmail, Google Calendar, Slack, Canva and Notion connections. Ten languages are supported, language selection is manual and the product remains in beta across platforms.[1], [2]
Model and task choice move into the voice layer
OpenAI asks users to choose Work or Codex first; a live transcript supports review while actions stay within that experience's permissions. Anthropic makes model choice more visible and carries requests into calendar, email or document tasks through connected apps. OpenAI therefore emphasizes one voice request advancing several tasks, while Anthropic exposes model and application selection. Which routing mechanism works better remains uncertain; neither announcement reports time saved, completion or error rates.[1], [2]
Convenience does not erase constraints
Anthropic's update did not change the underlying voice model, and free users remain limited to Haiku and one app connection. OpenAI offers desktop Voice to eligible accounts according to plan and workspace permissions but gives no Codex-only weekly-user count. Stronger model or connection choices do not mean basic speech behavior was rebuilt. The signal to watch is independent measurement of completion, correction, unauthorized actions, latency and cost under the same tasks and permissions.[1], [2]
Related columns
For more information on this topic, you can read the related columns.
