OpenAI has released GPT-6 Astra, a model designed to operate inside existing business software without requiring custom integrations or workflow redesign. OpenAI confirmed in its GPT-6 Astra product announcement that the model is now available in ChatGPT Work, Codex, and the API.
What GPT-6 Astra Does Differently in Business Environments
Most enterprise AI deployments require organizations to restructure data pipelines, rebuild workflows, or develop API integrations before the technology becomes productive. GPT-6 Astra is designed to bypass that requirement. According to OpenAI's announcement, Astra can operate within applications that have no API at all, writing code and executing tasks through the same desktop interfaces that employees already use.
OpenAI cited its own internal rollout as evidence of that capability. Engineering teams used Astra to identify and resolve a memory-allocation bottleneck in Codex test environments. By switching memory allocators based on Astra's findings, those teams achieved 25 times lower turn latency alongside approximately 30 percent higher peak memory use, according to the same announcement.
On the creative and marketing side, OpenAI stated that its developer and marketing teams used Astra and Codex to convert three hours of multicamera footage into a finished video production, a task it described as taking roughly three hours of human effort with the model's assistance.
Performance on Professional and Financial Tasks
OpenAI benchmarked Astra against the Financial Modelling World Cup, citing the 2023 Microsoft Excel World Championship as the reference event. According to OpenAI's announcement, Astra completed those challenges approximately four times faster than the winning human competitor using computer use capabilities.
On Terminal-Bench 4.0, a benchmark that tests agents on software engineering, system configuration, and data analysis tasks, Astra reached 57.9 percent accuracy. OpenAI's announcement states that figure compares to 37.3 percent for GPT-5.6 Sol and 55.8 percent for Claude Fable 5.1. OpenAI also states that Astra achieves this at approximately 9 percent lower estimated API cost per task than GPT-5.6 Sol, and approximately 63 percent lower than Claude Fable 5.1.
Serena Ge, Co-Founder and CEO of Datacurve, is quoted in OpenAI's announcement stating that Astra "sets a new record on DeepSWE v1.1 at 74%" and that it did so "with fewer steps and greater token efficiency than has ever been achieved by frontier models."
API pricing is set at $10 per million input tokens and $50 per million output tokens, as stated in OpenAI's announcement.
Brand Adherence and Document Output Quality
OpenAI's announcement emphasizes that Astra is designed to follow a company's existing voice, templates, and design standards more consistently than prior models. The practical implication, as stated by OpenAI, is that initial outputs require fewer rounds of revision before they meet business standards. The announcement demonstrates this with a slide deck example in which Astra produced a formatted presentation using only a few reference slides from an existing OpenAI template.
Silas Alberti, SVP of Research at Cognition, is quoted in OpenAI's announcement describing Astra's computer use, writing, and codebase understanding as improving output quality "right out of the box," with videos described as "noticeably easier to follow" and reports as "clearer and more concise."
For marketing and content teams, the combination of brand adherence, computer use, and document generation capabilities makes Astra a candidate for tasks that have historically required significant human review cycles. Evaluating the model's practical value in those workflows would reasonably involve measuring review time per deliverable, the number of revision rounds before approval, and cost per usable output, metrics that reflect actual productivity gains rather than benchmark scores alone.
Enterprise Controls and Safety Designations
OpenAI's announcement states that enterprise administrators can restrict Astra's access to approved websites and desktop applications, control file uploads and downloads, and manage browsing history. ChatGPT Work and Codex also include confirmation policies that require human approval before the model takes consequential actions, along with automated review of tool calls flagged as potentially unsafe or unauthorized.
OpenAI also disclosed that GPT-6 Astra is the first model to reach the "Critical" cybersecurity capability threshold under the company's Preparedness Framework. In response to that designation, OpenAI states it has strengthened safeguards against misuse, improved the model's resistance to attempts to bypass safety boundaries, and deployed automated checks designed to block harmful outputs.
On the data side, Zero Data Retention is available for eligible API customers on supported endpoints, subject to OpenAI approval.
New enterprise plugins for ChatGPT Desktop are launching alongside Astra. OpenAI's announcement lists integrations with Oracle Analytics, Power BI (a Microsoft Fabric service), Navan, and Avalara as the initial plugin partners.
Availability
GPT-6 Astra is available now through ChatGPT Work, Codex, and the API. Enterprise access is off by default at launch, and administrators must enable the model under their applicable rate card and agreement, according to OpenAI's product announcement.
