en.Wedoany.com Reported - OpenAI launched GPT-Live and ChatGPT Work this week, expanding its artificial intelligence ecosystem capabilities. GPT-Live is a next-generation voice model that will become the default mode for ChatGPT Voice, adopting a full-duplex architecture that allows simultaneous listening and speaking, eliminating the artificial pauses present in previous versions.
Traditional voice models convert speech to text, process the response, and then convert it back to audio, while GPT-Live processes speech in real time, deciding when to listen, respond, or remain silent. For complex questions, the system calls upon a backend reasoning model (GPT-5.5 at launch) to maintain conversational flow. Users can choose from three reasoning levels: Instant, Medium, and High. New features include visual cards displaying weather forecasts, stock data, and sports information during conversations, as well as simultaneous interpretation. The model can also remain silent when users request a delayed response. GPT-Live-1 becomes the default voice model for Go, Plus, and Pro plans, with a lightweight version serving the free plan, available on iOS, Android, and ChatGPT.com. OpenAI states that over 150 million people use ChatGPT's voice features weekly.
OpenAI also launched ChatGPT Work, an agent that aggregates user application and workflow data to create materials such as spreadsheets, slides, documents, and web applications, and can handle long-duration tasks by breaking complex projects into smaller steps. Built on the programming tool Codex, ChatGPT Work integrates a new desktop application, combining Chat, Work, and Codex in one place, and connects to services like Google Drive, Slack, and Salesforce through a unified plugin directory. Access begins with Pro, Enterprise, and Edu plans, with Plus and Business users gaining access a few days later. The desktop application (for Mac and Windows) has been released globally for all plans, including the free plan. To mitigate risks, the company created the Auto-Review system, which checks before accessing data or using connected services; OpenAI stated that in internal security tests, the system blocked all attempts to extract protected data.
ChatGPT Work is based on the GPT-5.6 series, comprising three models: Sol (most advanced), Terra (cost-effective mid-range option), and Luna (most affordable choice). OpenAI defines it as a next-generation product for programming, research, cybersecurity, professional work, and AI agent usage. In the API, Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra at $2.50 and $15; Luna at $1 and $6. The series underwent a testing phase limited to selected partners before launch. This release intensifies competition with Anthropic: analysts describe ChatGPT Work as a direct competitor to Anthropic's recently launched Claude Cowork. Sol outperforms some competing models in tests but lags behind Claude Mythos 5 in software development capabilities. By combining more natural voice conversations with agents capable of executing tasks in applications, files, and browsers, OpenAI reinforces its commitment to a new phase of AI: systems that not only answer questions but also help complete daily activities.
