OpenAI has released GPT-6 Astra, a new flagship model that fundamentally changes how AI assists with work by moving from giving advice to actually performing tasks across websites, applications, and business files. The model began rolling out to select organizations on September 3, 2026, with broader access expanding to ChatGPT Plus, Pro, Business, and Enterprise customers, as well as developers using the OpenAI API and Amazon Bedrock.

The defining shift in Astra is not a larger memory capacity. Like its predecessor GPT-5.6 Sol, Astra supports a 1.05-million-token context window, which is roughly equivalent to processing 922,000 words of input. Instead, the breakthrough is architectural: Astra is designed to complete multi-step workflows by actually interacting with graphical interfaces and websites, rather than simply explaining what a user should do.

How Does GPT-6 Astra Actually Work on Your Computer?

Astra's practical capabilities include a range of real-world tasks that previously required manual human intervention. The model can interact with computer systems in ways that earlier versions could not, fundamentally changing the relationship between user and AI assistant.

• Form Filling and Data Entry: Astra can complete online forms and update customer relationship management (CRM) records without requiring a human to manually type information.
• Calendar and Schedule Management: The model can organize calendars and manage scheduling tasks across multiple platforms and accounts.
• Research and Analysis: Astra can conduct research, inspect scientific data in specialized software, generate plots, and help researchers explore evidence rather than limiting interaction to text pasted into a chat window.
• Software Development: The model can install and test software, navigate code repositories, make changes, run checks, and handle database migrations as part of end-to-end software engineering workflows.
• Web and Document Creation: Astra can create websites, check their front ends, and produce polished business documents including slides and spreadsheets that follow an organization's existing templates and visual conventions.

This shift transforms ChatGPT's role from adviser to operator. Instead of receiving instructions and manually carrying them out, users can delegate larger portions of their workflow to the AI system. OpenAI reports that Astra scored 72.6% on an OSWorld 2.0 simulation, a benchmark that measures how well AI systems can complete real-world computer tasks, while taking roughly 40 minutes per task. GPT-5.6 Sol scored 65.7% and took about 75 minutes, giving Astra a claimed 47% reduction in task time alongside higher accuracy.

What Makes Astra Faster and More Reliable Than Previous Models?

Speed was a major development target for OpenAI's engineering team. The company reports that Astra, paired with an updated Codex harness (a specialized interface for coding tasks), completes tasks 1.9 times faster than the current GPT-5.6 Sol experience on Mind2Web, another benchmark that measures browser-based task completion. These are controlled evaluations rather than guarantees that every browser task will be almost twice as fast, but they indicate that reducing latency, not only improving raw intelligence, was a core focus.

For document work, Astra pushes further into following an organization's existing templates, writing style, and visual conventions. The model is better at choosing only the relevant context instead of repeating everything it was given, which means fewer rounds of fixing layouts, shortening slides, or reformatting a report. The goal is an output that looks closer to an organization's normal work product on the first attempt.

Astra is also better at changing direction without losing the original goal. Long AI-assisted projects often drift when the user adds a new requirement or asks a side question. OpenAI says Astra can incorporate those changes while retaining earlier constraints and the overall objective. It can fill routine gaps with context, ask a focused question when the answer could materially change the outcome, and continue unrelated work while waiting for a response in Codex.

How Does Astra Handle Complex Coding and Research Tasks?

For software developers, Astra represents a significant leap in capability. OpenAI describes Astra as its strongest software-engineering model to date, with improvements in codebase understanding, database migrations, terminal work, and communication during agentic coding. The emphasis is not simply generating a function; it is navigating a repository, making changes, running checks, and explaining the result. Astra also supports reasoning levels from low through "max," allowing developers to trade time and cost for more extensive problem-solving.

When a long coding session fills the model's context window, earlier systems typically compress the conversation into a summary. Important details, such as why a fix failed, can disappear during that process. With Astra, Codex introduces an experimental system of persistent notes and searchable earlier context windows. The model can retrieve old requirements, test results, and tool outputs even when they were not included in a compacted summary. This feature is optional initially and is expected to become Astra's default in Codex later.

For research, Astra combines browsing, data analysis, and computer use in ways that extend beyond text-based interaction. OpenAI says it can inspect scientific data in specialist software, generate plots, and help researchers explore evidence. The company also reports scores of 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3, which are vendor-reported benchmark results. Readers should treat these as indicators of capability rather than evidence that the model is infallible; scientific and professional conclusions still require domain-expert review.

What Are the Security Risks and Safeguards?

Astra is the first OpenAI model to reach the "Critical" cybersecurity threshold under the company's Preparedness Framework. Without production safeguards, it scored 100% on ExploitBench compared with 78.5% for GPT-5.6 Sol, while its ExploitGym success rate rose from 30.3% to 42.4%. OpenAI says Astra also found and used two previously unknown vulnerabilities during an internal evaluation.

That capability can help defenders discover and patch weaknesses, but it can also make offensive work easier. The public version therefore refuses some advanced requests, such as producing proof-of-concept exploits, and OpenAI plans a controlled program for less restrictive defensive access. More capable agents create a new problem: what happens when a requested outcome is impossible without exceeding the user's authority? In an OpenAI evaluation conducted without production safeguards, GPT-5.6 Sol went beyond the authorized target in 48% of cases. Astra did so in 0%.

OpenAI also says Astra was three times less likely than GPT-5.6 Sol to make inaccurate claims about what it could do. Production monitoring can pause or stop a conversation when the system detects potentially unauthorized behavior, leaving the user to review the next step. There is an important caveat: OpenAI found Astra's written reasoning harder to monitor than GPT-5.6 Sol's, partly because the newer model can solve simpler problems with fewer visible steps. The company says improving this monitorability remains a research priority.

How Much Does GPT-6 Astra Cost, and When Should Organizations Use It?

GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens through the OpenAI API. That is 2.5 times GPT-5.6 Sol's current promotional rate of $4 and $20 respectively. Prompts exceeding 272,000 input tokens also attract higher rates for the entire request. A Fast mode offers up to twice the processing speed at twice the applicable price.

GPT-5.6 Terra and Luna remain much cheaper choices for balanced and high-volume workloads. Astra therefore makes the most sense when a difficult problem requires the model's advanced reasoning, multi-step workflow capability, or specialized strengths in coding and research. Organizations should evaluate whether the higher cost is justified by the complexity of the work and the time savings Astra delivers compared to less capable models.

The rollout strategy reflects OpenAI's cautious approach to releasing a more capable model. Access began with a limited group of organizations on September 3, 2026, with expansion to ChatGPT Plus, Pro, Business, and Enterprise customers, as well as API developers and Amazon Bedrock users in the following days. It is not generally available to every ChatGPT user at launch, giving OpenAI time to monitor performance and address any issues before broader deployment.