GPT-6 Astra AGI: What Is OpenAI’s New AI Model and How Powerful Is It?

OpenAI has introduced GPT-6 Astra, describing it as a new generation of intelligence designed for advanced reasoning, computer use, coding, scientific research, cybersecurity and professional work.

The model represents a major step forward in AI capabilities, with OpenAI reporting significant improvements across several demanding benchmarks. However, while searches around “GPT-6 Astra AGI” are likely to attract considerable interest, OpenAI’s announcement does not explicitly describe Astra as Artificial General Intelligence (AGI).

What Is GPT-6 Astra?

GPT-6 Astra is OpenAI’s latest frontier AI model. According to OpenAI, it combines advances in pre-training, reinforcement learning and alignment to handle complex real-world tasks.

The company says Astra is particularly strong in areas including computer use, software engineering, cybersecurity, science and professional knowledge work.

OpenAI reports that GPT-6 Astra achieved a 99.9% score on ARC-AGI-3, a 98% score on FrontierMath Tier 4, and a 100% score on ExploitBench in its reported evaluations.

Is GPT-6 Astra AGI?

The term AGI, or Artificial General Intelligence, generally refers to an AI system capable of performing a broad range of intellectual tasks at a highly general level rather than being optimized for a narrow application.

GPT-6 Astra demonstrates capabilities across a remarkably broad collection of tasks. It can work with software, browse and interact with computers, perform research, analyze scientific information, write code and create professional documents.

However, the OpenAI announcement itself does not state that GPT-6 Astra is AGI. Therefore, calling Astra “AGI” as an established fact would go beyond what the source confirms.

A more accurate description is that GPT-6 Astra is a highly capable frontier AI model with broad, general-purpose capabilities.

GPT-6 Astra Benchmark Performance

OpenAI’s reported benchmark results show substantial gains over previous models.

The results cover abstract reasoning, mathematics, science, coding and cybersecurity.

GPT-6 Astra for Coding

Coding is one of the major areas where Astra is positioned as a significant improvement.

OpenAI describes GPT-6 Astra as its best model for software engineering to date. Its reported 57.9% score on Terminal-Bench 4.0 compares with 37.3% for GPT-5.6 Sol in the same table.

Astra also scored 74.1% on DeepSWE and 63.9% on internal database migration tasks.

The model is designed not only to generate code but also to work through complex development tasks, inspect existing codebases and perform testing and troubleshooting.

GPT-6 Astra Computer Use

Computer interaction is another major focus of Astra.

OpenAI says the model can perform tasks such as filling online forms, updating customer records, organizing calendars, researching information and working inside document and spreadsheet applications.

It can also create websites, generate plots, analyze data and perform frontend quality-assurance checks.

On OSWorld 2.0, OpenAI reports a 72.6% score for Astra, compared with 65.7% for GPT-5.6 Sol in the comparison presented.

GPT-6 Astra and Cybersecurity

Cybersecurity is one of the most notable areas of Astra’s capability.

OpenAI says Astra reached a 100% score on ExploitBench in testing without production safeguards, compared with 78.5% for GPT-5.6 Sol. It also achieved 42.4% on ExploitGym and 88.0% on SRE-Bench.

OpenAI also says its testing identified situations where Astra could discover and use previously unknown vulnerabilities. The company states that two such vulnerabilities discovered during evaluation were disclosed to their maintainers.

At launch, however, Astra is designed to refuse certain advanced cybersecurity requests, including creating proof-of-concept exploits for vulnerabilities.

GPT-6 Astra for Science and Mathematics

Astra is also designed for scientific and mathematical work.

OpenAI reports a 96.0% score on GPQA Diamond, a benchmark covering graduate-level questions in biology, chemistry and physics.

The model achieved 97.6% on FrontierMath Tier 4, according to OpenAI’s benchmark table.
OpenAI also says Astra helped with research concerning prime-number gaps, including work that established a stronger bound for infinitely many pairs of primes and improved a bound concerning unusually large gaps between primes.

GPT-6 Astra for Professional Work

Beyond coding and research, Astra is designed for professional knowledge work.

OpenAI says the model can create and work with documents, presentations, spreadsheets and analyses while following existing templates and visual styles.

It is also designed to make better decisions when instructions are incomplete. Astra can use context to fill routine gaps while asking focused questions when missing information could materially change the outcome.

GPT-6 Astra vs GPT-5.6

OpenAI’s reported results show Astra outperforming GPT-5.6 Sol across several benchmarks.

For example:

• ARC-AGI-3: 99.9% vs 7.8%
• Terminal-Bench 4.0: 57.9% vs 37.3%
• GPQA Diamond: 96.0% vs 94.6%
• ExploitBench: 100% vs 78.5%
• SRE-Bench: 88.0% vs 55.9%
• ARC-AGI-2: 95.0% vs 92.5%

These figures come from OpenAI’s published evaluation table and represent the maximum reported scores at the evaluated effort levels.

GPT-6 Astra Availability

OpenAI says GPT-6 Astra is initially rolling out to a limited set of organizations before becoming available to ChatGPT Plus, Pro, Business and Enterprise users.

The model is also being made available through the OpenAI API, Microsoft Azure and AWS Bedrock.

For developers, OpenAI lists the API model identifier as gpt-6-astra.

OpenAI lists standard API pricing at $10 per million input tokens and $50 per million output tokens, with separate pricing for cached tokens.

What Makes GPT-6 Astra Different?

The biggest change is not simply that Astra can answer questions more accurately. OpenAI is positioning it as an AI system that can reason, use computers and execute multi-step workflows.

That combination could make the model more useful for tasks that previously required a person to move information between multiple applications.

Astra can potentially research information, work with software, analyze data, create documents and perform other connected tasks rather than simply returning a text response.

GPT-6 Astra and the Future of AGI

GPT-6 Astra’s capabilities will inevitably fuel discussion about whether modern frontier models are approaching AGI.

Its performance across mathematics, coding, computer use, science, cybersecurity and professional work demonstrates a much broader capability range than traditional narrow AI systems.

But AGI is not determined by a single benchmark score. It involves broader questions about generalization, autonomy, reliability, adaptability and performance across the full range of human intellectual activities.

For now, the most accurate conclusion is that GPT-6 Astra represents a major advance toward increasingly general-purpose AI, but the supplied OpenAI announcement does not officially declare Astra to be AGI.

Bottom Line

GPT-6 Astra is OpenAI’s new frontier AI model focused on advanced reasoning, computer use, coding, science, cybersecurity and professional workflows.

Its reported benchmark performance shows substantial gains over GPT-5.6 Sol in several areas, including abstract reasoning, coding and cybersecurity.

While “GPT-6 Astra AGI” is likely to become a popular search phrase, it is important to distinguish speculation about AGI from OpenAI’s actual announcement. Astra is clearly presented as a highly capable general-purpose AI model, but the source does not officially label it Artificial General Intelligence.