OpenAI is betting on agents. On September 10, 2026, the company introduced a new Data agent in ChatGPT Work, giving organizations a way to analyze connected company data and documents. The announcement arrives one week after OpenAI announced GPT-6 Astra, a frontier model the company says may mark the beginning of artificial generalized intelligence.
The Data agent connects to approved sources including Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake. It can also bring files and documents from Google Drive and SharePoint into an analysis, so the data does not need to remain trapped in one tidy warehouse — a condition most companies have never experienced.
Organizations can give the agent their own business terms, metric definitions, custom calculations, and data relationships. Enterprise administrators choose which data connections are available and which roles can use them, while queries enforce the connected account’s existing table-, row-, and column-level restrictions.
After analyzing the information, the Data agent can turn its work into an interactive dashboard with built-in visualizations. That makes the feature less about producing another answer in a chat window and more about creating a reusable view of company performance.
GPT-6 Astra makes the larger claim
On September 3, 2026, OpenAI announced GPT-6 Astra and framed it as a possible starting point for the AGI era. Greg Brockman, OpenAI co-founder and president, opened with: “Welcome to the AGI era.”
Astra is designed to navigate software much as a person does, working across browsers, spreadsheets, websites, and desktop applications. It can fill out online forms, update CRM records, organize calendars, conduct web research, draft results into documents or email, manipulate spreadsheets, and analyze scientific data in Python notebooks.
The model can also work in Power BI, create and test websites, operate engineering applications such as KiCad and FreeCAD, and install and troubleshoot software. OpenAI says Astra handled tasks ranging from creating a 3D game to preparing a legal agreement while managing unrelated requests.
Astra begins rolling out Thursday to enterprise customers through OpenAI’s gated access program, Daybreak. Over the coming days, it will become available to ChatGPT Plus, Pro, Business, and Enterprise customers, as well as through the OpenAI API and cloud platforms including AWS Bedrock and Microsoft Azure.
On an offline subset of OSWorld 2.0, Astra scored 72.6% while taking roughly 40 minutes per task. GPT-5.6 Sol scored 65.7% and took roughly 75 minutes, giving Astra the stronger result in both task score and completion time.
The model also scored 97.6% on FrontierMath Tier 4 v2, 74.1% on DeepSWE v1.1, 95.9% on BenchCAD, 96% on GPQA Diamond, 100% on ExploitBench, and 98.6% on ARC-AGI-3. OpenAI describes Astra as its largest-scale training run, pretrained using more than 100,000 DBUs at Stargate infrastructure, with its capabilities attributed to large-scale pretraining and reinforcement learning.
AGI remains a moving target
OpenAI’s AGI claim comes with an obvious complication: nobody agrees on the finish line. Brockman said, “Everyone has a different definition of AGI,” and described the idea as less like a single historic moment than a “gray, fuzzy thing.”
He also said, “I do think we’re there,” referring to Astra qualifying as AGI. Later, Brockman said, “I think it’s not unreasonable to feel that we are now in the AGI era.” The language carries confidence, but it also admits that the label depends on the definition.
The benchmark picture adds another layer. NVIDIA’s AVO architecture achieved a 100% score across all 25 environments and 183 levels in the ARC-AGI-3 public set, using Claude Opus 5 with a baseline of roughly 30%. NVIDIA’s AVO adds persistent memory, tools, feedback, and recovery, while OpenAI’s Astra uses the company’s Responses API; comparison models can operate under different configurations.
That makes benchmark victories useful but incomplete. Astra’s case rests on broad software control, demanding research and engineering tasks, strong scores, and a large training run — not on one magic number that settles the argument.
The Data agent shows the more immediate business consequence. OpenAI is pairing systems that can reason across software with systems that can interpret governed company data, apply internal definitions, respect permissions, and produce dashboards. AGI may remain a debate over terminology; enterprise automation has already been given a login screen.
Based on
