OpenAI Launches Data Agent for ChatGPT Work and GPT-Live-1 Voice API

OpenAI introduced a Data agent for ChatGPT Work that allows users to interact with company data, create dashboards, and take actions through natural language prompts. [1] The agent connects to existing data repositories, including Snowflake, Databricks, BigQuery, MongoDB, and Redshift, as well as file systems like Google Drive and SharePoint. [1]

It integrates with business intelligence platforms such as Tableau, Power BI, Sigma, and ThoughtSpot, enabling users to edit, share, and refresh dashboards. [1] Administrators manage data connections, and the system respects existing table, row, and column permissions while incorporating established company metrics and calculations. [1] Users can share their findings via email or Slack and perform approved actions using the `@Data` command. [1]

OpenAI also launched GPT-Live-1 in the API, a model that enables full-duplex voice conversations capable of handling pauses and interruptions. [2] While GPT-Live-1 manages the live conversation, it delegates deeper reasoning tasks and actions to a backend model or agent framework. [2]

When paired with GPT-6 Astra at a medium reasoning effort level, the model completed 83.6 percent of Tau3 support tasks on the first attempt, compared to 45.7 percent for GPT-Realtime-2.1. [2] The model reduces turn-taking latency to 0.798 seconds, down from 1.41 seconds in the previous GPT-Realtime-2.1 model. [2] The release includes 12 real-time voices and supports connections via WebRTC, WebSockets, or SIP for telephony agents. [2]

OpenAI is expanding its enterprise and developer offerings by addressing both backend data analysis and frontend customer interaction simultaneously. The Data agent positions ChatGPT Work as a direct competitor to specialized BI interfaces by integrating natively with tools like Tableau and Power BI. Meanwhile, GPT-Live-1's latency reduction to sub-second levels and its dramatic performance improvement on the Tau3 benchmark over GPT-Realtime-2.1 indicate a push toward production-ready voice agents for telecom and retail support. By separating the real-time voice orchestration from the deep reasoning of models like GPT-6 Astra, OpenAI allows developers to mix and match capabilities. What the sources don't address: How pricing scales for GPT-Live-1 when continuously delegating reasoning tasks to heavier backend models like GPT-6 Astra.