Without you noticing: ChatGPT is changing
The weekly upgrade from OpenAI and Anthropic proves that voice is becoming the primary interface. With the new GPT-Live models and access to source code, the ChatGPT app is no longer just talking — it is managing desktop workflows.
The era of typing and text-based interfaces is preparing to recalculate its route: artificial intelligence giants OpenAI and Anthropic (the developer of Claude) have announced a series of significant updates to their voice modes. This time, it is a dramatic change — from a user talking to a mobile screen to a "personal assistant" that actually manages desktop applications on a computer.
The most important step comes from OpenAI, which has launched a major update to the ChatGPT desktop application (on Windows and macOS systems). The update integrates the new GPT-Live family of voice models and allows users to conduct a continuous and fluent voice dialogue while activating artificial intelligence agents to perform complex actions on a personal computer.
Not just talk
While in smartphone versions we have become accustomed mainly to fluid conversations without the ability for deep execution, the new desktop update changes the picture completely. ChatGPT Voice is now fully integrated into the advanced work environments of ChatGPT Work and Codex (the company's software development environment) and knows how to take command of complex multi-stage tasks.
For macOS users, the system is capable of examining the active window on the screen with a click or a voice command ("Take a look at this") and analyzing the visuals or the Alt-text appearing in it. In the company's demonstration, a developer was shown asking ChatGPT, by voice only, to open a new thread, perform a Pull Request in the code network, and find the root cause of a bug in the system — all with one continuous voice command.
In addition, iOS app users can now remotely control the Codex running on their desktop computer using voice commands.
How does it work in practice? Unlike old-generation voice generators, GPT-Live models are capable of understanding speech fragments, interrupting answers mid-sentence if necessary, and identifying when the user needs to provide additional input to complete a task on the computer.
Anthropic's response
In parallel with OpenAI's announcement, the prominent competitor Anthropic is also not remaining idle and has upgraded the voice mode of its Claude. If until now Claude's voice mode was based mainly on the light Haiku model, now users can run voice conversations also based on Sonnet and Opus — the company's most powerful and intelligent models.
The main advantage of Anthropic in this round lies in the deep connection to third-party applications: Claude's voice mode is capable of accessing work applications such as Gmail, Google Calendar, Slack, Notion, and Canva directly during the conversation and performing complex tasks in them directly from the voice dialogue.
