OpenAI releases GPT-6.2 Astra flagship model for complex tasks
The new model targets high-end engineering and scientific research with specialized visual reasoning and a large 1.05 million token context window.
The short version
- GPT-6.2 Astra is designed for hard projects like scientific research, real-world engineering, and 3D designs.
- The model features a 1,050,000 token context window and supports up to 128,000 completion tokens.
- OpenAI reports the model achieves new evaluation highs at less than a third of the cost per task compared to GPT-6 Astra.
- Pricing is set at $8.00 per 1M input tokens and $40.00 per 1M output tokens via OpenRouter.
Why it matters
This release signals a shift toward specialized, high-reasoning models for professional industries rather than general-purpose chat. It provides a tool for researchers and engineers who need to process massive amounts of data in a single prompt.
A flagship for complex work
OpenAI has introduced GPT-6.2 Astra as the premier model in its latest series. This flagship version is specifically built to handle the most difficult projects currently facing users. It is intended for professional environments where accuracy and reasoning are paramount.
The model is tailored for heavy-duty applications such as real-world engineering and scientific research. It also shows improved visual reasoning capabilities. These improvements make it particularly useful for creating websites, slideshows, and 3D designs.
According to openrouter.ai, the model was released on November 17, 2026. It represents a step toward more specialized AI tools that can manage complex, multi-step instructions. The model is designed to be the go-to option for the hardest tasks in the GPT-6.2 lineup.
Technical specs and limits
One of the most notable technical features of GPT-6.2 Astra is its massive context window. It can process up to 1,050,000 tokens in a single request. This allows users to feed the model very large documents or extensive datasets for analysis.
In addition to the large input capacity, the model supports up to 128,000 completion tokens. It also accepts various file types as input, including PDFs, images, and standard text files. While it processes these diverse inputs, it currently returns text as the output format.
The model supports advanced features like tool calling and structured outputs via a JSON schema. These capabilities allow developers to integrate the model into existing software workflows more easily. It can also perform web searches, which are billed at a specific rate of $10.00 per 1K calls.
Pricing and performance
The pricing for GPT-6.2 Astra is structured based on input and output volume. The standard rates are $8.00 per 1 million input tokens and $40.00 per 1 million output tokens. There are also specific costs for different types of data interactions, such as cache reads and writes.
OpenAI notes that the model achieves new highs on its evaluations at a lower price point than previous versions. Specifically, it costs less than a third of the price per task of the GPT-6 Astra model. This suggests a focus on making high-performance reasoning more economically viable for large-scale use.
Users can also benefit from automatic prompt caching for inputs of 1024 tokens or more. OpenAI bills a cache write on each cached prompt regardless of whether the prefix is reused. This system helps manage costs when users frequently send similar large blocks of information to the model.
Availability and routing
Currently, GPT-6.2 Astra is served by one primary provider on OpenRouter, which is OpenAI itself. Because it is a flagship model, it is not used in the standard routing modes. Instead, users must choose specific routing modes to access it based on their needs.
OpenRouter offers different modes like Balanced, which balances price and speed, and Nitro, which focuses on the fastest response. There is also a Floor mode for users who want the cheapest available option. These modes help users navigate the different price points offered by various endpoints.
The model is not currently available through in-region routing for certain areas like the EU. However, OpenRouter provides tools to monitor provider uptime and availability. This ensures that if one provider has an error, the system can potentially route requests to a healthy provider if the user's filters allow it.
Developer integration
For developers, GPT-6.2 Astra is designed to be easily integrated into existing applications. OpenRouter provides API drop-in code that is compatible with the OpenAI standard. This means most existing SDKs will work with the new model by simply swapping the model slug.
The model supports tool_choice for function calling, which is a standard way for AI to interact with external software. By using a JSON schema in the response_format, developers can ensure the model returns data in a predictable structure. This is critical for building reliable apps that need to parse AI responses.
While the model is powerful, it is part of a larger ecosystem of OpenAI models. Other versions like GPT-6.2 Astra Pro and Sol Pro exist alongside it. Developers can choose the specific version that best fits their balance of performance requirements and budget constraints.
What to watch for
The release of GPT-6.2 Astra marks a significant milestone in the move toward high-reasoning AI. It is important to see how professional engineering and scientific firms adopt these capabilities in their daily workflows. The model's ability to handle 3D designs and visual reasoning will be a key area to watch.
Another factor to monitor is the cost-to-performance ratio as more providers potentially host the model. As more companies join the infrastructure, the 'Floor' and 'Balanced' pricing might become more competitive. This could lead to wider adoption in industries that were previously priced out of high-end AI.
Finally, the effectiveness of the 1.05 million token context window in real-world applications remains to be seen. Users will need to test how well the model retains information over such a long sequence. This will determine if it truly becomes the standard for large-scale data analysis.
Try it
Check the OpenRouter documentation to see the specific model slug and API requirements for integrating GPT-6.2 Astra into your own code.
Sources and links
First reported by openrouter.ai. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.