OpenAI Launches GPT-6 Astra With Major Advances in AI Agents and Coding

OpenAI has introduced GPT-6 Astra, its latest frontier AI model, with a focus that goes beyond answering questions or generating text. The company says Astra is built to handle complex, multi-step tasks across computers, software development, research and professional workflows, allowing users to hand over more of the work itself to an AI system. The model is initially rolling out to a limited group of organisations, with broader access planned through ChatGPT Plus, Pro, Business and Enterprise plans, as well as the OpenAI API. Microsoft Azure and AWS Bedrock will also get access as the rollout expands.

GPT-6 Astra is built to work on a computer

One of the biggest changes with GPT-6 Astra is its ability to interact with computers rather than simply tell people how to perform a task. OpenAI says Astra can fill out online forms, update records in customer relationship management systems, organise calendars and carry out online research. It can also work inside document editors and spreadsheets, analyse data, create websites and test whether those websites function properly.

That makes the model more useful for tasks where the difficult part isn't coming up with an answer but actually getting a series of actions done. OpenAI's testing also points to a significant improvement in speed. In latency simulations using OSWorld 2.0, Astra scored 72.6%, compared with 65.7% for GPT-5.6 Sol, while taking roughly 40 minutes per task instead of about 75 minutes. OpenAI says that translates to about 47% less time per task in the test.

Coding and professional work get a bigger push

GPT-6 Astra is also being positioned as OpenAI's strongest model for software engineering so far. The model can work through codebases, make changes and handle longer sequences of development tasks. OpenAI has also updated the Codex harness alongside Astra, saying the combination can complete certain computer-use tasks 1.9 times faster than the previous GPT-5.6 Sol experience on the Mind2Web benchmark.

The company is targeting more than coding. Astra has been trained for professional work that involves documents, spreadsheets and presentations, including following existing templates and adapting outputs to a user's instructions and visual style. OpenAI says the model is also better at deciding what information actually matters instead of simply filling an output with everything it has found. That could be particularly useful in areas where AI-generated work still needs to be checked by a professional.

OpenAI is putting more weight on AI agents

The broader change behind GPT-6 Astra is the move towards AI agents that can complete workflows rather than individual tasks. Astra can take a goal, work through several steps and adjust when requirements change. OpenAI says its testing found the model stayed within an authorised target in all cases in one evaluation designed around lessons from the company's earlier Hugging Face security incident. GPT-5.6 Sol, without production safeguards, went beyond the authorised target in 48% of those tests. That doesn't mean the model can simply be left unsupervised for every task. OpenAI is still putting access and safety controls around the system, particularly because its capabilities are considerably stronger in areas such as cybersecurity.

Cybersecurity is both a strength and a concern

GPT-6 Astra is the first OpenAI model to reach the Critical level of cybersecurity capability under the company's Preparedness Framework. OpenAI says Astra can, with the right tools and access, discover previously unknown vulnerabilities and develop ways to exploit them across well-protected systems without a person guiding every step. Because of that capability, the company has added stronger safeguards designed to prevent harmful cyber actions resulting from either misuse or model behaviour. At the same time, OpenAI says Astra is more resistant to jailbreak than its predecessors, following additional safety training and extensive internal and external testing.

From answering questions to doing the work

The arrival of GPT-6 Astra reflects where OpenAI is taking its models next. The emphasis is increasingly on systems that can reason, use software, navigate the web and complete workflows, rather than stopping after producing a response. For users, that could mean asking an AI system to research a topic, organise the findings, prepare a document and carry out related computer tasks in one workflow. For businesses, the potential is even broader, with software agents increasingly able to work alongside existing tools.

GPT-6 Astra is only beginning its wider rollout, but its launch makes one thing clear: OpenAI is trying to make its newest models less like assistants that simply provide answers and more like systems that can take responsibility for completing complicated pieces of work.