OpenAI launches GPT-6.2 Astra and Sol for GitHub Copilot
New models offer developers more efficient coding, better security, and improved performance on complex multi-step software tasks.

The short version
- GPT-6.2 Astra is optimized for advanced coding, scientific research, and terminal-based tasks.
- GPT-6.2 Sol is designed for intricate agentic workflows and comprehensive code reviews at a lower cost.
- Astra scored 5.2 percentage points higher than its predecessor on the Agents' Last Exam benchmark.
- OpenAI reduced input and output token rates for the Astra model API by 20% compared to GPT-6 Astra.
- Both models include enhanced cybersecurity features and meet critical thresholds for safety monitoring.
Why it matters
For developers and tech companies, these models mean more precise defect detection and the ability to handle complex software projects with fewer errors. It also provides a way to perform high-level reasoning tasks without significantly increasing the price of running the AI.
New models for developers
OpenAI has officially released two new models, GPT-6.2 Astra and GPT-6.2 Sol, which are now being integrated into GitHub Copilot. These models are designed to help software engineers perform more complex tasks, such as identifying specific defects in code and managing multi-step development projects. As reported by Mint, the rollout targets users of GitHub Copilot Pro, Max, Business, and Enterprise plans.
The update aims to make the coding process more precise by allowing developers to pinpoint errors more effectively. By leveraging these new models, teams can handle intricate software development tasks that require deeper reasoning than previous versions could provide. The integration ensures that the AI can better understand the nuances of professional programming environments.
GitHub confirmed the availability of these models in its official changelog on November 17. They are being deployed across a wide range of popular development environments, including Visual Studio Code, Visual Studio, JetBrains IDEs, and Xcode. This wide availability ensures that a broad spectrum of professional developers can access the new capabilities immediately.
Astra vs. Sol specializations
While both models represent a step forward, they are built for different professional use cases. GPT-6.2 Astra is specifically optimized for advanced coding and terminal-based tasks, which are common in software development and scientific research. It is also designed to create well-structured documents, spreadsheets, and presentations while adhering to established formats.
In contrast, GPT-6.2 Sol is geared toward intricate agentic workflows and comprehensive code reviews. Agentic workflows refer to tasks where the AI can perform a series of actions to reach a goal, rather than just answering a single prompt. Sol is intended to be a more efficient model for general technical tasks that require high-level reasoning without adding to operational costs.
The performance gap between the two models highlights their different priorities. For instance, OpenAI noted that GPT-6.2 Sol beat GPT-6.2 Astra by only 0.5 points on Cognition's FrontierCode 2.0 test, yet the Sol model is roughly 20 times cheaper. This allows businesses to choose between the high-power capabilities of Astra or the cost-effective reasoning of Sol depending on their specific needs.
Benchmarks and performance
OpenAI provided several benchmarks to demonstrate the improvements over previous generations. GPT-6.2 Astra scored 5.2 percentage points higher than the previous GPT-6 Astra on the Agents' Last Exam, which measures how well an AI handles multi-step professional tasks. It also set a new record on the Terminal-Bench-Science 0.1, showing a 9-point improvement over its predecessor.
Beyond raw coding power, the models show improved performance in initial code review tests. GitHub stated that these new models are more capable of analyzing existing codebases to find bugs or suggest improvements. These improvements are particularly useful for large-scale software projects where manual review can be time-consuming and prone to human error.
The models also show promise in real-world, everyday tasks. GPT-6.2 Sol outperformed the previous GPT-6 Astra on OpenAI's LifeVal evaluation. This suggests that while Astra handles the heavy lifting of scientific and technical coding, Sol remains highly competent at understanding and executing common, everyday instructions provided by users.
Pricing and security updates
Cost efficiency is a major component of this release, particularly for enterprise users. OpenAI reduced the input and output token rates for the GPT-6.2 Astra model API by 20% compared to the GPT-6 Astra model. This helps organizations scale their AI usage while managing their budgets more effectively as they integrate these tools into their workflows.
The pricing for GPT-6.2 Sol remains the same as its predecessor, GPT-6.1 Sol. It costs $2 per million input tokens and $10 per million output tokens. This consistency allows companies to switch to the newer Sol model for its improved reasoning and coding capabilities without having to recalculate their existing operational expenses.
Security remains a primary focus for both new models. OpenAI's system card indicates that both Astra and Sol meet critical thresholds for cybersecurity and high thresholds for biological and chemical capabilities. The company has implemented additional restrictions and monitoring to minimize the risk of misuse, ensuring that the models remain safe for professional use in sensitive environments.
Rollout and availability
The transition to these new models will not happen all at once for every user. GitHub has stated that the rollout will be gradual and will occur on a case-by-case basis. This means that while the models are technically available, some organizations may need to work with their GitHub business or enterprise administrators to enable them.
The availability of these models across various platforms ensures that developers can use them in their preferred tools. Whether a developer prefers the Xcode environment for mobile apps or JetBrains IDEs for enterprise software, the new GPT-6.2 models will be accessible. This flexibility is key for maintaining developer productivity during the transition to more advanced AI tools.
Looking ahead, the focus for these models will likely be on further refining agentic behaviors. As developers begin to use Astra for complex 3D design and gaming, and Sol for deep code reviews, OpenAI will likely gather data on how these models perform in diverse professional settings. The goal is to continue refining the balance between high-performance reasoning and cost-effective execution.
What to watch
- How developers prioritize Astra over Sol for specific project types.
- Future updates to the GPT-6.2 system card regarding safety thresholds.
Try it
If you are a GitHub Copilot subscriber, check your settings or contact your administrator to see if the GPT-6.2 Astra or Sol models are available for your account.
Sources and links
- ReportingMint: the article we sourced
First reported by Mint. 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.