OpenAI has released GPT-6 Astra, its most capable model yet for coding, research, computer use, document creation, and complex multi-step workflows. The launch moves Astra from a closely watched safety milestone into an actual product—and gives builders a clearer view of what OpenAI believes the next flagship model should do.
Astra is built for work that extends beyond producing an answer. In the Responses API it can search the web, retrieve files, run code, operate computers, generate images, use a hosted shell, apply software patches, connect through MCP, and discover tools when needed. Its 1.05-million-token context window and 128,000-token maximum output are aimed at jobs that span large codebases, document collections, and long execution histories.
The flagship model becomes an operator
The most important shift is not a single benchmark score. Astra is positioned around end-to-end work: understanding an objective, selecting tools, maintaining context, adapting when requirements change, checking intermediate results, and finishing a usable artifact.
In ChatGPT, OpenAI says Astra can create documents, spreadsheets, and presentations that follow existing templates and instructions. For developers, the supported tool set makes the model a foundation for coding agents, research systems, workplace automation, and applications that must act across several environments rather than remain inside a chat window.
That makes the surrounding agentic harness inseparable from model performance. Permissions, memory, tool routing, recovery, evaluation, and human approval determine whether intelligence becomes reliable work—or simply a more capable source of unpredictable actions.
Early deployments test complete workflows
OpenAI's launch examples focus on operational outcomes rather than short demonstrations. Legal platform Legora used an Astra-powered agent to review 41 documents in minutes during a financial-statement workflow. The company says the system found all four planted errors, including a hidden £500,000 discrepancy, and improved nearly 40% over the previous model on that workflow.
Game-development platform Playco used Astra inside an AI-powered development environment connected to Unity and Godot. The system could edit scenes, play and test games, find bugs, and improve what it built. Playco reports 50% fewer manual fixes than with the previous model and says most of three themed prototypes worked on the first attempt.
These are vendor-reported results, not independent benchmarks, but they demonstrate the intended product direction: agents that ingest substantial context, operate inside professional tools, and produce a reviewable result in one continuous run.
Access starts narrow and expensive
Astra is rolling out first to enterprises in OpenAI's Trusted Access Program. OpenAI says API access and availability through ChatGPT Plus, Pro, Business, and Enterprise plans will follow in the coming days, so the model is not yet generally available to every user.
Standard API pricing is $10 per million input tokens, $1 per million cached input tokens, and $50 per million output tokens. Prompts above 272,000 input tokens carry higher rates for the full request. Those economics position Astra for high-value tasks where stronger completion rates, fewer failed iterations, or reduced manual review can justify flagship pricing.
More autonomy requires stronger boundaries
Astra is also OpenAI's first model classified at the Critical level for cybersecurity capability. As we covered in our earlier analysis of Astra's security threshold, OpenAI says the model can—with suitable tools and access—find previously unknown vulnerabilities and develop exploits across well-defended systems without a person directing every step.
The release adds monitoring to all tool-using Astra inference. ChatGPT may pause or stop a conversation when monitoring suggests that an agent misunderstood its instructions, allowing the user to review the situation before work continues. OpenAI also reports that Astra is more resistant to prompt injection and less likely than GPT-5.6 Sol to take unauthorized or destructive actions in simulated workplace environments.
There is an important tension. OpenAI says Astra is better aligned overall, yet can be harder to monitor through its chain of thought under adversarial conditions. That makes external controls—least-privilege access, approval gates, audit trails, isolated environments, and reversible actions—more important as models become more capable operators.
Why it matters for builders
GPT-6 Astra signals that flagship-model competition is moving from isolated intelligence toward reliable completion of entire workflows. The key question is no longer only whether a model can write code, analyze a document, or call a tool. It is whether the system can preserve context, coordinate actions, revise its approach, verify the result, and stop when human judgment is required.
For product teams, the opportunity is to design around outcomes rather than prompts. A specialized agent that completes one valuable workflow safely and consistently can be more useful than a general assistant with an impressive model but weak permissions, state management, and verification.
Astra raises the ceiling for those products. It also raises the standard for the harness surrounding them.