GPT-6 Astra Launch: Features, Price, AGI Claims and Availability

OpenAI has officially launched GPT-6 Astra, its newest frontier artificial intelligence model, and the company is presenting it as a major step forward in AI capability, computer use and professional work.

The new model was unveiled on September 3, 2026, and is now beginning its wider rollout. OpenAI describes GPT-6 Astra as its “most intelligent and aligned model”, with significant improvements in reasoning, software engineering, browsing, computer use, scientific research and complex multi-step tasks.

What makes GPT-6 Astra particularly notable is that it is not designed simply to answer questions faster. OpenAI says the model can actually operate computers, navigate websites, complete professional workflows, create software and work with specialised tools.

The launch has also reignited the debate around Artificial General Intelligence (AGI), with some technology leaders suggesting that models such as Astra could represent the beginning of a new phase in AI development.

What Is GPT-6 Astra?

GPT-6 Astra is OpenAI's latest generation of frontier AI technology. It combines improvements in pre-training, reinforcement learning and alignment to handle tasks that previously required significant human intervention.

According to OpenAI, Astra is particularly strong at computer use, browsing, software development, cybersecurity, mathematics, science and professional knowledge work.

Instead of simply generating instructions for a user, the model can perform multi-step tasks itself.

For example, Astra can work with websites, organise information, update records, create documents and presentations, analyse data and even test software.

That makes the model particularly important for the growing AI agent market.

GPT-6 Astra Can Use Computers and Browsers

One of the biggest changes with Astra is its computer-use capability.

OpenAI says the model can handle tasks such as filling online forms, updating CRM records, organising calendars and conducting online research. It can also work inside document and spreadsheet applications.

The company reports that Astra scored 72.6% on OSWorld 2.0, compared with 65.7% for GPT-5.6 Sol in the comparison published by OpenAI.

OpenAI also reports a major efficiency improvement. In its latency simulations, Astra achieved higher computer-use performance in around 47% less time per task than GPT-5.6 Sol.

This could have a major impact on businesses where employees currently spend hours performing repetitive digital tasks.

GPT-6 Astra Shows Huge Benchmark Gains

OpenAI has highlighted several benchmark results to demonstrate Astra's capabilities.

The model reportedly achieved:

• 99.9% on ARC-AGI-3
• 98.5% on ARC-AGI-1
• 95% on ARC-AGI-2
• 96% on GPQA Diamond
• 95.9% on BenchCAD
• 91.5% on BrowseComp
• 57.9% on Terminal-Bench 4.0

OpenAI also reports a 100% score on ExploitBench, a cybersecurity evaluation testing whether AI systems can develop working exploits from known vulnerabilities.

These numbers are based on OpenAI's own evaluation methodology, so they should not automatically be interpreted as proof that Astra is better than every competing model in every real-world situation.

Still, the results demonstrate how quickly frontier AI systems are progressing.

GPT-6 Astra and the AGI Debate

Perhaps the biggest story surrounding GPT-6 Astra is the question of whether increasingly capable AI systems are approaching AGI.

OpenAI has positioned Astra as a major step toward systems capable of performing a broad range of economically valuable intellectual work.

NVIDIA CEO Jensen Huang also made headlines following the launch by saying that “AGI has arrived”, although this is his assessment rather than an independent scientific declaration that GPT-6 Astra definitively meets a universally accepted definition of AGI.

Experts remain divided because there is no single universally accepted test for AGI.

A system can outperform humans on benchmarks while still having limitations in reliability, autonomy, common-sense reasoning and understanding of the physical world.

GPT-6 Astra Is Also Much Stronger at Professional Work

OpenAI is positioning Astra not only as a chatbot but as a tool for professional workflows.

The model can reportedly create and modify documents, presentations, spreadsheets, websites, applications and analyses while following existing templates and organisational requirements.

For companies, this could mean using AI for tasks that previously required multiple applications and human operators.

Astra can also make decisions when instructions are incomplete. OpenAI says the model is better at recognising when it should make a reasonable assumption and when it should stop and ask the user for clarification.

This distinction could become increasingly important as AI agents gain more control over real-world digital systems.

GPT-6 Astra Gets Major Cybersecurity Capabilities

The cybersecurity side of GPT-6 Astra is attracting considerable attention.

OpenAI says Astra reached the Critical threshold for cybersecurity under its Preparedness Framework. The model demonstrated significantly stronger capabilities at identifying and developing exploits than previous models.

On OpenAI's ExploitBench evaluation, Astra achieved 100%, compared with 78.5% for GPT-5.6 Sol.

These capabilities could be useful to security researchers trying to discover vulnerabilities before criminals exploit them.

However, they also create new risks. A more capable AI system could potentially make sophisticated cyber attacks easier to develop.

That is why OpenAI says additional safeguards and monitoring are an important part of Astra's deployment.

GPT-6 Astra Price

OpenAI has announced an API price of $10 per million input tokens and $50 per million output tokens for GPT-6 Astra's standard processing. Cache reads and writes have separate rates.

A faster processing mode is also available, with OpenAI saying Fast mode can deliver up to twice the speed at twice the standard price.

For regular ChatGPT users, access is being rolled out through existing subscription plans rather than requiring a separate Astra subscription.

GPT-6 Astra Availability

OpenAI says GPT-6 Astra initially began rolling out to a limited group of organisations before expanding to:

• ChatGPT Plus
• ChatGPT Pro
• ChatGPT Business
• ChatGPT Enterprise
• OpenAI API
• Microsoft Azure
• Amazon Bedrock

Pro, Business and Enterprise users are also set to receive access to GPT-6 Astra Pro. Enterprise administrators can control whether Astra is enabled for their workspace.

The rollout has not been completely smooth, however. Reports on September 7 said some paying users experienced difficulties accessing the new model during the rollout, prompting an apology from OpenAI CEO Sam Altman.

Why GPT-6 Astra Matters

GPT-6 Astra could represent an important shift in how people use AI.

Earlier generations primarily focused on generating answers. Astra is increasingly focused on doing the work.

That distinction is significant.

A user could potentially ask an AI agent to research information, work across websites, prepare a spreadsheet, create a presentation and check the final result rather than manually completing every step.

For students, developers, researchers, designers and businesses, that could dramatically change everyday workflows.

At the same time, the greater autonomy of AI creates difficult questions around security, accountability and human oversight. Reuters reported that concerns about monitoring increasingly capable agents are already part of the discussion surrounding Astra's launch.

Final Thoughts

GPT-6 Astra is much more than another incremental chatbot upgrade. OpenAI is presenting it as a major leap toward AI systems capable of reasoning, using computers, writing software and completing complicated professional tasks with much less human intervention.

Its benchmark results are impressive, while its computer-use and cybersecurity capabilities could have particularly large real-world consequences.

Whether GPT-6 Astra should actually be called AGI remains open to debate. But there is little doubt that the launch marks another major step in the race toward increasingly autonomous AI systems.