OpenAI deploys GPT-6 Astra to ChatGPT Work, Codex and API
On September 9, OpenAI announced the enterprise deployment of the GPT-6 Astra model for ChatGPT Work, Codex and API. The company cites a million-token context window and its own assessment of critical cyber capabilities.
GPT-6 Astra for businesses: On September 9, OpenAI announced the enterprise deployment of the GPT-6 Astra model for ChatGPT Work, Codex and API. The company presents the model as a tool for workplace and enterprise use, including agentic workflows that can work with websites and desktop applications.
OpenAI first announced the model on September 3. A new company page titled “GPT-6 Astra: The next generation in intelligence for work” describes its deployment in products designed for working with code, chat and programmatic interfaces.
GPT-6 Astra for businesses in API
In OpenAI’s documentation, the model is listed under the identifier gpt-6-astra. It has a context window of 1.05 million tokens and a maximum output length of 128,000 tokens. API pricing is set at $10 per million input tokens and $50 per million output tokens.
A context window of this size could be relevant when processing extensive documents, longer work histories or large code repositories. However, the parameters alone do not indicate what performance the model will achieve on a specific enterprise task.
OpenAI says Astra can work with websites and desktop applications in relevant environments, including where an API is not available. The model is therefore not intended only to create text or code proposals, but also to perform steps in the work tools being used.
Approval of high-risk actions
For application control, the company describes approval rules for high-risk actions and enterprise access restrictions. This is an important part of deploying systems that can affect workflows instead of merely providing an answer to the user.
OpenAI also classifies GPT-6 Astra in its system card as a model with critical cyber capability under its own Preparedness Framework. This designation is OpenAI’s assessment under its internal methodology, not an independent regulatory decision.
In its materials, the company describes safeguards and restrictions on access to advanced capabilities. The specific practical scope of availability to users and businesses will need to be monitored as the individual products are deployed in practice.
What has not been independently verified so far
OpenAI presents performance data, comparisons and customer references on its own websites or through partner organizations. The editorial team has not independently verified the cited benchmarks or the claimed savings from using the model.
Likewise, the specific technical architecture of the model is not confirmed in the primary materials reviewed. Interpretations concerning so-called looped transformers have appeared in external commentary, but OpenAI has not confirmed them in its basic materials.
For GPT-6 Astra for businesses, it will therefore be important to monitor independent tests, audits of security measures, customer experiences beyond marketing references and any details about the limits of its cyber functions.
Sources
- OpenAI — GPT-6 Astra: The next generation in intelligence for work – Confirms the enterprise launch for ChatGPT Work, Codex and API, enterprise controls and the stated ability to work in applications.
- OpenAI — GPT-6 Astra: A new generation of intelligence – Confirms the model’s announcement on September 3, planned availability channels and stated evaluation results.
- OpenAI Developers — GPT-6 Astra model – Confirms the API model name, context window, maximum output length, pricing data and supported capabilities.
- OpenAI Deployment Safety Hub — GPT-6 Astra System Card – Documents OpenAI’s own assessment that the model reaches the critical cyber capability threshold, as well as the safeguards applied.
Verified and updated: 09/10/2026 06:24
