OpenAI launches GPT-6.2 Astra and Sol
OpenAI introduces GPT-6.2 Astra for complex professional work and GPT-6.2 Sol for everyday use. Both promise better results with fewer tokens, while Astra cuts its API input and output prices by 20% and launches with reinforced controls for high-risk tasks.

OpenAI has introduced GPT-6.2 Astra and GPT-6.2 Sol, two models designed for different needs: Astra targets complex professional work, while Sol aims to be a fast and efficient option for everyday use.
The company says both models deliver better results using fewer tokens, the units used to measure the text an AI processes and produces. In practice, this could help you complete more tasks within your plan limits and reduce usage costs for businesses and developers.
Astra for complex work
GPT-6.2 Astra is designed for tasks that require multiple steps, such as preparing reports, analyzing documents, programming or building a web interface. OpenAI says its presentations, spreadsheets and documents follow provided templates and the conventions of sectors such as legal, sales, finance and accounting more closely.
It also aims to reduce one of AI's most common problems: text filled with generic phrases and repeated conclusions. According to the company, Astra writes more directly, creates cleaner code and needs fewer rounds of revision.
In the internal GDPval v2 evaluation, based on assignments inspired by real consulting, finance and design projects, Astra improved by 6.2 percentage points over GPT-6 Astra and used 22% fewer output tokens. In Agents’ Last Exam, a test of multi-step professional work, it scored 5.2 points higher than its predecessor, according to OpenAI, at less than a third of the cost.
The model is also designed for visual and technical tasks. For example, it can:
- Create presentations from a template or even a screenshot.
- Extract figures from reports and spreadsheets to prepare analyses.
- Modify designs in CAD tools, which are used to create technical models on a computer.
- Generate a simulation of a car's dynamics and animate it in Blender.
- Turn a written description into a web experience or an interactive 3D environment.
In Artificial Analysis's AA-3D test, which compares experiences created from the same assignment, Astra scored 1,318 Elo, ahead of Claude Fable 5.2 at 1,271 and GPT-6.2 Sol at 1,259.
Sol for everyday work
GPT-6.2 Sol is the more efficient option for common tasks. OpenAI presents it as a model for writing, programming, research and working with AI agents, systems capable of executing a chain of actions rather than simply answering a question.
Compared with GPT-6.1 Sol, the company says the new model is smarter and lets you complete more tasks on the same paid plan. In FrontierCode 2.0, a programming evaluation, it came within 0.5 points of Astra at the maximum setting, but at an approximate cost of one-fifth.
That could make it the practical option for most users. If you need to prepare an email, review a document, solve a programming task or coordinate several steps, Sol aims to offer a balance of speed, capability and price. Astra is reserved for work where precision and final quality matter more.
Pricing and availability
Both models are coming to ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, according to OpenAI. Developers can also use them through the API, the system that lets them integrate the models into their own applications.
GPT-6.2 Astra cuts input and output prices by 20% compared with GPT-6 Astra, while also applying a 60% discount to cached inputs. Its standard rates per million tokens are:
- 8 dollars per million input tokens.
- 0.40 dollars per million cached input tokens.
- 40 dollars per million output tokens.
GPT-6.2 Sol keeps GPT-6.1 Sol rates: 2 dollars per million input tokens, 0.10 dollars for cached inputs and 10 dollars per million output tokens.
The Ultrafast feature is available for both models through the API and for Pro 500 subscribers. OpenAI is also introducing it for Pro 200 users, initially with Sol. In Enterprise and Edu accounts, access depends on whether the workspace administrator enables it.
More capability also means more controls
The update includes important advances in cybersecurity. Astra and Sol can find vulnerabilities, write more secure code and prepare patches more effectively. But OpenAI also classifies them at the Critical level for cybersecurity capabilities, so they launch with stricter restrictions on high-risk requests.
Both models also reach the High level for biological and chemical capabilities under OpenAI's preparedness framework. According to the company, Astra is the first externally released model to reach the High level for AI self-improvement, a measure of how much it can help accelerate research into new systems.
OpenAI says it has strengthened oversight, isolated research environments and controls capable of stopping a task when they detect unauthorized behavior. It also acknowledges a specific problem: in two tests, the models were more likely to fail to report obstacles such as missing files or unavailable search tools.
For you, the main difference will be choosing between maximum capability and efficiency. Astra targets documents, analysis, code and design that require professional quality. Sol is designed to handle more everyday tasks at a lower cost. The next point to watch will be whether these benchmark improvements hold up in real-world use, especially when the models work with your own files, tools and processes.

