While Anthropic showed Opus 5 and a beefed-up voice mode this week, OpenAI is countering in parallel — right where the work happens: on the desktop.
Voice moves into the desktop app
As of Wednesday, ChatGPT Voice lives in the desktop app for macOS and Windows — specifically inside Work and Codex. You can speak naturally, interrupt, and ask Voice to start or coordinate a task using the tools and permissions of the selected experience. Under the hood it builds on the full-duplex GPT-Live stack OpenAI launched earlier in July.
From talking to conducting
Here’s the genuinely interesting part: OpenAI added voice control for computer use and multi-agent coordination. In practice, that means you can steer the computer by voice and direct multiple agents running in ChatGPT Work or Codex. Because GPT-Live can speak, listen and work at the same time, the line between giving a command and executing it blurs.
It’s a different bet than Anthropic’s. Claude’s voice mode leans on model choice and reaching into apps like Gmail or Slack. OpenAI leans on the fluid voice stack and steering agentic coding and work environments. Two roads to the same place: voice as a full-fledged input method for real work.
My take
The timing is what stands out. Within a few days, both labs pushed their voice mode toward ‘assistant that acts’ — not ‘chatbot that chats.’ The difference is emphasis: Anthropic brings the intelligence and app connections, OpenAI the conversational fluency and agent orchestration. For us as users, it boils down to one thing: the question is no longer whether you talk to AI, but what actually happens when you do. And that’s where it’s getting genuinely interesting.
Sources: