On September 3, OpenAI launched GPT-6 Astra, bringing what it describes as the next major step for ChatGPT. Astra is not just designed to answer questions or generate content. It can handle multi-step tasks, use a computer, work with software, conduct research, write code and create documents, spreadsheets and presentations.
OpenAI calls Astra its “most intelligent and aligned” model and says it could mark the beginning of the AGI era. The bigger change is how ChatGPT is being positioned. Instead of simply telling users what to do, Astra is built to increasingly do the work itself, moving ChatGPT closer to an AI agent that can operate across real-world digital tasks.
What Is GPT-6 Astra
GPT-6 Astra is OpenAI’s latest flagship model, built for complex, end-to-end work rather than just answering prompts. It can interact with computers, browsers and software, conduct research, write and test code, work with business documents and handle multi-step tasks with less human intervention.
OpenAI describes it as its most capable model for tasks that require reasoning, tool use and execution across different environments.
Astra is built to handle a wide range of work, including:
- Complex reasoning for difficult problems that require multiple steps and decisions.
- Coding including writing, debugging and working with larger codebases.
- Computer use such as navigating websites, filling forms and working inside software.
- Research across the web and large amounts of information.
- Software engineering including development, testing and troubleshooting.
- Scientific work such as analyzing data and generating plots.
- Professional document creation including documents, spreadsheets and presentations that follow existing templates and instructions.
GPT-6 Astra Key Specifications
- Context window: 1.05 million tokens
- Maximum output: 128,000 tokens
- Knowledge cutoff: April 30, 2026
- Reasoning levels: Low, medium, high, xhigh and max
- Model ID:
gpt-6-astra
The 1.05 million-token context window allows Astra to work with very large amounts of information in a single task, while the 128,000-token output limit gives it room to produce lengthy responses or work products when needed. Its April 30, 2026 knowledge cutoff also gives it a more recent base of knowledge than OpenAI’s previous GPT-5.6 models.
GPT-6 Astra Benchmarks Show Large Gains Over GPT-5.6
GPT-6 Astra shows major gains over GPT-5.6 Sol across reasoning, coding, computer use and automation benchmarks. The biggest jump comes in agentic tasks.
| Benchmark | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|
| ARC-AGI-3 | 99.9% | 7.8% |
| FrontierMath Tier 4 | 97.6% | 83.0% |
| GPQA Diamond | 96.0% | 94.6% |
| OSWorld 2.0 | 72.6% | 65.7% |
| ScreenSpot-Pro | 92.7% | 76.9% |
| Terminal-Bench 4.0 | 57.9% | 37.3% |
| AutomationBench | 41.4% | 18.1% |
| BenchCAD | 95.9% | 83.3% |
| BrowseComp | 91.5% | 90.4% |
What GPT-6 Astra Costs
Because GPT-6 Astra is designed for more demanding, end-to-end tasks, its API pricing is higher than GPT-5.6 Sol. OpenAI charges $10 per 1 million input tokens and $50 per 1 million output tokens for Astra.
For the API:
- Input: $10 per 1 million tokens
- Cached input: $1 per 1 million tokens
- Output: $50 per 1 million tokens
- Requests above 272K input tokens: charged at higher rates, with input and cache rates doubled and output priced at 1.5×
- Fast mode: costs 2× the applicable standard rates
By comparison, GPT-5.6 Sol costs $4 per 1 million input tokens and $20 per 1 million output tokens. Its cached input rate is $0.40 per million tokens.
This means Astra costs 2.5× more than GPT-5.6 Sol for both standard input and output. The higher price reflects Astra’s positioning as a model for complex reasoning, coding, computer use and other high-value workloads rather than simple, high-volume queries.
GPT-6 Astra Shows How Fast the AI Race Is Moving
GPT-6 Astra is a major step forward, but its launch also shows why more capable AI brings bigger risks. OpenAI classified Astra as its first model to reach the Critical level of cybersecurity capability, meaning it can identify previously unknown vulnerabilities and develop exploits against well-protected systems with the right tools and access.
OpenAI says it added stronger safeguards, monitoring and access controls before release, while advanced cybersecurity capabilities will initially remain more restricted.
The launch has also attracted controversy beyond security. Astra’s increasing ability to work autonomously has raised questions about how well its actions can be monitored, with OpenAI itself acknowledging that Astra’s monitorability has decreased compared with GPT-5.6 Sol in some adversarial tests.
The rollout also faced criticism after paying ChatGPT users were initially unable to access the model as expected, leading Sam Altman to apologize for what he called a “messy rollout.”
And this is what makes the current AI race so difficult to predict. Every few weeks, the leading models are gaining new abilities in reasoning, coding, computer use and autonomous task execution. A model that looks ahead of the competition today can quickly be challenged by another release.
With OpenAI, Anthropic and Google all pushing toward more capable AI agents, predicting what these systems will be able to do even a few months from now is becoming harder. For now, GPT-6 Astra is another clear sign that the race is moving beyond better chatbots and toward AI systems that can increasingly do the work themselves.
