OpenAI released GPT-6 Astra on September 3, 2026. The company describes Astra as its most capable model for complex reasoning, coding, computer use, research, and document creation. The official name is GPT-6 Astra, rather than ChatGPT-6, because the same model is designed to work across ChatGPT, Codex, and the OpenAI API.
The release began as a staged rollout. OpenAI said early access was reaching enterprises in its Trusted Access Program, with access through the API and Plus, Pro, Business, and Enterprise plans following. That makes the announcement a product launch, but not a promise of immediate availability for every account.
A model designed to finish complete workflows
Astra's central idea is broader than a better response in a chat window. OpenAI positions it for work that spans several steps, tools, applications, and checks before a finished result is ready. Its published strengths include browsing, software engineering, science, professional work, and computer use.
The model specification lists a 1,050,000-token context window, up to 128,000 output tokens, and a knowledge cutoff of April 30, 2026. It accepts text and images and produces text. The large context can help with long documents and codebases, but teams still need retrieval, validation, and clear task boundaries because context size alone does not guarantee a correct result.
Long-running work gets better controls
Three new controls make the release especially relevant to agent products. Async tool calling lets Astra continue useful work while an application runs a function or custom tool. Mid-turn steering lets a person send a correction while a response is still in progress. Configuration updates can also change reasoning effort during a conversation without rebuilding the cached prompt prefix.
These are interface decisions as much as model features. A useful agent experience needs to show what is running, what is waiting, what changed after new instructions, and which result still needs review. Better steering reduces the pressure to choose between interrupting a task and letting the wrong plan continue.
Developers get more capability and new constraints
Developers can call the model with the identifier gpt-6-astra. It supports low, medium, high, xhigh, and max reasoning effort, but not the none setting. OpenAI also says custom temperature, top_p, and log-probability settings are unsupported. Applications using tools need the Responses API even though the model can also answer through Chat Completions.
At launch, OpenAI listed text pricing at $10 per million input tokens, $1 per million cached input tokens, and $50 per million output tokens. Cache writes were listed at $12.50 per million tokens, with higher rates for prompts above 272,000 input tokens. Teams evaluating Astra should therefore measure cost per completed task, including tool calls and retries, rather than comparing token prices in isolation.
Access depends on the product and workspace
ChatGPT Learn says Astra appears in the model selector once it is available to an account. Enterprise access also requires an eligible rollout and an administrator to enable the model. API access is governed separately by the API organization and project connected to a key.
That distinction matters for product teams planning a launch around the model. A feature can be technically ready while customers, employees, or test environments still have different access. Availability checks, fallbacks, and clear messaging belong in the release plan.
More capable agents raise the bar for trust
OpenAI says supported Astra requests can be checked asynchronously for signs of misalignment, with monitoring able to raise alerts or stop work for review. The safeguard reflects the operational change created by models that can browse, write code, use computers, and act across several systems.
Products built around this capability need narrow permissions, visible approval points, durable logs, and a clear way to inspect evidence. The strongest experience will feel less like an all-knowing chatbot and more like a supervised workbench where people can understand the plan, steer the process, and verify the final result.
What the release means for design teams
GPT-6 Astra moves the design challenge from composing a single prompt toward shaping an entire working relationship. Interfaces need to communicate progress without noise, preserve context across corrections, and keep consequential actions legible before they happen.
For designers and engineers, the opportunity is not simply to add Astra to an existing chat box. It is to design workflows where the model's new reach is matched by stronger feedback, control, recovery, and accountability.