OpenAI has introduced GPT-6 Astra, a new flagship artificial intelligence model designed to handle complex reasoning, coding, computer-use tasks, scientific work and professional workflows.
The company describes Astra as its most intelligent and aligned model so far. According to OpenAI, the model combines advances in pre-training, reinforcement learning and alignment to improve its ability to perform multi-step tasks and work with computers and software.
GPT-6 Astra focuses on computer use and professional tasks
One of the key areas of improvement is computer use. OpenAI says GPT-6 Astra can perform tasks such as filling online forms, updating records, organising calendars and conducting online research.
The model can also analyse scientific data, generate plots, create websites and perform frontend quality checks. Its capabilities are intended to allow it to complete longer workflows rather than simply generate text in response to individual prompts.
OpenAI reported a 72.6% score on OSWorld 2.0, compared with 65.7% for GPT-5.6 Sol in the comparison provided by the company. It also reported a 59.3% score on Agents’ Last Exam and 92.7% on ScreenSpot-Pro.
GPT-6 Astra brings stronger coding capabilities
OpenAI says GPT-6 Astra is its strongest model for software engineering to date. The model is designed to work on complex coding tasks and can use tools to inspect software, execute code and test applications.
Astra also introduces improved context handling in Codex. According to OpenAI, the model can preserve and retrieve information from earlier context windows, helping it maintain requirements and test results during longer coding sessions.
Advances in science and mathematics
GPT-6 Astra has also been developed for scientific and mathematical applications. OpenAI says it achieved a 96% score on GPQA Diamond, a benchmark covering graduate-level questions in biology, chemistry and physics.
The company also reported a 99.9% score on ARC-AGI-3 and a 98% score on FrontierMath Tier 4. OpenAI further said Astra helped establish new results related to both unusually small and unusually large gaps between prime numbers.
These results are benchmark claims made by OpenAI and should not be interpreted as meaning that the model has achieved human-level intelligence across every area.
Cybersecurity capabilities also see a major jump
Cybersecurity is another area where Astra represents a substantial increase in capability. OpenAI says the model meets the Critical threshold for cybersecurity under its Preparedness Framework.
In the company’s testing, Astra achieved a 100% score on ExploitBench, compared with 78.5% for GPT-5.6 Sol. OpenAI also said the model discovered and used two previously unknown zero-day vulnerabilities during one evaluation and that the vulnerabilities were disclosed to their maintainers.
Because of these capabilities, OpenAI says additional safeguards have been introduced to reduce the risk of harmful cyber use.
GPT-6 Astra rollout and availability
OpenAI says GPT-6 Astra is initially rolling out to a limited group of organisations. The company plans to make it available over the following days to ChatGPT Plus, Pro, Business and Enterprise users.
The model is also being made available through the OpenAI API and AWS. Developers can access it through the API using the model identifier gpt-6-astra.
OpenAI’s API documentation lists a 1.05-million-token context window and a maximum output of 128,000 tokens for GPT-6 Astra.
Does GPT-6 Astra mean AGI has arrived?
The launch has renewed discussion around artificial general intelligence (AGI) because of Astra’s performance across a wide range of tasks. However, OpenAI’s official announcement describes GPT-6 Astra as a highly capable frontier model and does not state that the company has declared AGI achieved.
Therefore, it would be more accurate to describe Astra as a major advancement in AI capabilities rather than definitively calling its launch the arrival of AGI.