OpenAI has made GPT‑6 Astra available in ChatGPT Work, Codex and API. The company said it developed the model for computer use, web browsing, professional work, software engineering, cybersecurity and science. Since Astra can also work with applications that do not have an API, businesses can use it in their existing workflows without extensive preparation or engineering work.
In its first days, customers used the model to optimize GPUs, identify inconsistencies in financial statements and prepare presentations in line with brand standards. OpenAI teams also turned three hours of multicamera footage into the GPT‑6 Astra Developer First Impressions video using Astra and Codex; the video received more than 550k views in four days. The engineering team fixed a memory allocation issue that was causing slowdowns in Codex sessions, reducing turnaround latency by 25×; peak memory usage increased by approximately 30%.
Astra was said to complete tasks with fewer tokens and retries, thereby reducing costs. Pricing starts at $10 per million input tokens and $50 per million output tokens. The model produced 89% fewer undesirable results than GPT‑5.6 Sol and 74.7% fewer than Claude Fable 5.1.
Organizations will be able to limit approved websites and desktop applications, uploads and downloads, and browsing history. ChatGPT Work and Codex include approval policies and automated review. Enterprise integrations for Oracle Analytics, Power BI, Navan and Avalara were also announced. Astra became the first model to reach the Critical cybersecurity capability threshold under the Preparedness Framework. Zero Data Retention is available to eligible API customers upon approval; enterprise access is initially disabled by default.
Why it matters
This development is relevant to businesses seeking automation without changing their existing software infrastructure, as it takes AI use beyond being limited to chat interfaces and brings it into computer- and web-based workflows. Its ability to work with applications that do not have an API expands its potential applications beyond custom integrations that technical teams can build; however, which tasks require human oversight remains dependent on organizations’ own approval policies. The fact that enterprise access is initially disabled by default and that the Zero Data Retention option requires approval shows that data management is not resolved automatically. Reaching the Critical cybersecurity capability threshold also puts the management of misuse risks on the agenda alongside the model’s capabilities. Although it offers advantages in terms of lower token usage and retry costs, the increase in peak memory usage remains a variable that organizations need to assess separately in their performance calculations.
Background
Astra is not a new name in the FikirPilot archive: we have published 5 reports mentioning the name in the past 90 days; the latest is dated 6 September 2026.