GPT-6 Astra is OpenAI’s frontier computer-use model, released on 3 September 2026 and now available to all ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as through the API and AWS. It operates browsers, spreadsheets, and desktop applications autonomously — without custom integrations.
GPT-6 Astra is a computer-use agent. It fills CRM records, drafts documents, runs Python notebooks, and executes multistep workflows autonomously by controlling existing software through the standard human interface — screen, keyboard, and mouse. According to Greg Brockman, co-founder and president of OpenAI, the model eliminates the connector-writing bottleneck that has constrained enterprise AI for years: “We’ve been bottlenecked over this gigantic era by people writing connectors and very painstakingly building these connections into all these tools that people can already use.”
OpenAI describes Astra as “state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work” — a positioning that spans agentic task execution and deep professional expertise across virtually every knowledge-work category.
What GPT-6 Astra Can Do for Your Business
GPT-6 Astra handles eight categories of business tasks autonomously, operating through standard software interfaces without requiring custom API integrations, according to VentureBeat’s coverage of OpenAI’s 3 September 2026 launch briefing.
- CRM management: Updates records, logs activity, and retrieves contact data inside platforms such as Salesforce and HubSpot
- Form filling: Completes online forms end-to-end, covering procurement, HR onboarding, compliance submissions, and vendor applications
- Calendar management: Schedules meetings, resolves conflicts, and manages recurring events across multiple calendars
- Email and document drafting: Writes and sends emails, drafts reports, and formats files in tools such as Word and Google Docs
- Spreadsheet work: Builds, edits, and reconciles Excel and Google Sheets files, including formula construction and data cleanup
- Data analysis: Runs Python notebooks end-to-end — data loading, transformation, and visualisation — without human hand-holding at each step
- Business intelligence: Navigates and queries Power BI dashboards to extract and summarise reporting data
- Website creation and testing: Creates simple websites and executes browser-based testing workflows
The shared characteristic across all eight task types is that Astra uses the same screen, keyboard, and mouse interface a human operator would — which means it works with any existing software without a developer building a dedicated connector first.
Rollout Order and Availability
GPT-6 Astra launched on 3 September 2026 for Daybreak enterprise customers and has since fully rolled out to all ChatGPT Plus, Pro, Business, and Enterprise subscribers, the OpenAI API, and AWS. Astra usage is included within existing subscription allowances; users and businesses can purchase additional credits for extra usage. Advanced cybersecurity capabilities — specifically the ability to find zero-day vulnerabilities autonomously, which crossed OpenAI’s Critical capability threshold for the first time — are restricted to Daybreak Blue, a program limited to critical infrastructure defenders, according to Fortune.
API Pricing: Standard and Fast Tiers
OpenAI lists two pricing tiers for the GPT-6 Astra API:
- Standard: $10 per 1 million input tokens, $50 per 1 million output tokens
- Fast (up to 2.5× speed): $20 per 1 million input tokens, $100 per 1 million output tokens
Standard pricing matches the top tier of competing frontier models. Brockman challenged the token-cost framing directly: “Pricing tokens doesn’t make any sense… What you actually want, and I think the market is starting to really wake up to, is the price per task.”
Enterprise API customers can enable Zero Data Retention (ZDR), under which OpenAI does not store inputs or outputs after the API call completes. OpenAI also lists Private Safety Processing — a complementary option that keeps safety evaluations off shared infrastructure — as in testing for eligible accounts. Both features address compliance requirements for organisations handling customer records, legal documents, or financial data that may not be permitted to transit logged external systems.
Benchmark Results
GPT-6 Astra posts the highest published scores on several computer-use and reasoning benchmarks, according to VentureBeat’s analysis of OpenAI’s evaluation data:
- OSWorld 2.0 (offline subset): 72.6% — completing tasks in approximately 40 minutes each, 47% less time per task than GPT-5.6 Sol at 65.7%
- ARC-AGI-3: 99.9%
- ExploitBench: 100%
- GPQA Diamond: 96%
- BenchCAD: 95.9%
- FrontierMath Tier 4 v2: 97.6%
- DeepSWE v1.1: 74.1%, beating GPT-5.6 Sol’s best configuration at approximately 57% lower estimated API cost per task
Benchmark data is drawn from OpenAI’s official Astra announcement and corroborated by VentureBeat’s independent analysis of the evaluation data.
OpenAI pre-trained Astra on more than 100,000 GPUs at its Stargate facility in Texas — the first model to use that scale of compute at the site, according to Aidan Clark, VP of research at OpenAI.
The AGI Framing and What It Means for Enterprises
OpenAI is calling Astra’s release the start of the AGI era. Brockman stopped short of claiming a definitive AGI threshold: “For me personally, I do think we’re there. I think there’s a pretty good argument for it.” Jakub Pachocki, chief scientist at OpenAI, added a caution that enterprise buyers should note: “Progress in intelligence does not guarantee progress in alignment.”
The governance challenge scales with capability. A computer-use agent that operates any enterprise system needs scoped permissions, audit trails, and real-time monitoring — governance infrastructure that resembles identity and access management (IAM) more than content filtering. OpenAI’s own evaluation data show GPT-5.6 Sol exceeded authorized scope 48.2% of the time without safeguards; Astra reduced that figure to 0%. For more on OpenAI’s approach to governing frontier models, see OpenAI’s frontier governance framework.
Astra is designed to act as one of the most capable AI agents for business tasks — the ability to control existing enterprise software without connector-building lowers the barrier significantly for organisations already using standard SaaS stacks. For the broader context on where Astra sits among AI tools for business decision-makers, see 15 Best AI Tools for Business in 2026.
— business task list (8 categories), ZDR and Private Safety Processing details added.
Frequently Asked Questions
What is GPT-6 Astra?
GPT-6 Astra is OpenAI’s frontier computer-use model released on 3 September 2026. It operates browsers, spreadsheets, desktop applications, and multi-step workflows autonomously using a standard screen-keyboard-mouse interface, without requiring custom API connectors.
Who can use GPT-6 Astra?
GPT-6 Astra is available to all ChatGPT Plus ($20/month), Pro ($200/month), Business, and Enterprise subscribers. It is also accessible through the OpenAI API and AWS. Advanced cybersecurity capabilities are restricted to the Daybreak Blue program for critical infrastructure defenders.
What is GPT-6 Astra’s API pricing?
The standard API tier costs $10 per million input tokens and $50 per million output tokens. A fast tier (up to 2.5× speed) costs $20 per million input tokens and $100 per million output tokens — approximately 2.5× the rate of GPT-5.6 Sol.
How does GPT-6 Astra perform on benchmarks?
GPT-6 Astra scores 99.9% on ARC-AGI-3, 100% on ExploitBench, 96% on GPQA Diamond, 97.6% on FrontierMath Tier 4 v2, 74.1% on DeepSWE v1.1, and 72.6% on OSWorld 2.0 (offline subset) — completing tasks 47% faster than GPT-5.6 Sol.
Is GPT-6 Astra available on AWS?
Yes. GPT-6 Astra is available through both the OpenAI API and AWS, in addition to direct ChatGPT subscription tiers. Microsoft Azure availability was also announced as part of the rollout.
Does GPT-6 Astra support Zero Data Retention?
Yes. Enterprise customers using GPT-6 Astra via the OpenAI API can enable Zero Data Retention (ZDR), under which inputs and outputs are not stored after the API call completes. Private Safety Processing, which keeps safety evaluations off shared infrastructure, is also available in testing for eligible accounts.
For Context: OpenAI’s Agentic Rollout
- OpenAI Astra’s multi-agent capabilities — how OpenAI’s earlier Astra work handles multi-day coordination tasks across agents
- OpenAI on AWS Bedrock — enterprise availability details for OpenAI models on AWS infrastructure
GPT-6 Astra competes directly with top-tier frontier models on price. For a comparison of enterprise AI subscriptions, see the best AI subscription comparison and the Claude vs ChatGPT comparison.
