In brief
- GPT-6.2 Astra is the company’s new model for complex professional work; GPT-6.2 Sol is its option for everyday tasks.
- OpenAI says Astra can work in greater detail with components in CAD and simulation tools.
- In the company’s example, Astra created a 3D model of a car chassis, ran a simulation with a Python solver, and animated the results in Blender.
- In OpenAI’s AA-3D evaluation, Astra ranks first with an Elo score of 1,318; Sol scored 1,259.
OpenAI announced the GPT-6.2 Astra and GPT-6.2 Sol models on 17 November 2026. The company says Astra was developed for complex professional work, while Sol is positioned as an efficient option for everyday use. The announcement highlights precise interaction with CAD and simulation tools, 3D animation with Blender, and interactive 3D experiences that run in a browser.
What can Astra do in CAD and simulation workflows?
According to OpenAI, GPT-6.2 Astra can work with individual parts and connections in CAD and simulation tools. The company says this could help engineers examine how components fit together, see how loads change as a design moves, and test revisions before prototyping.
In one example from OpenAI, Astra created a 3D model to visualize the chassis dynamics of a car traveling on a mountain road. It then wrote a Python solver to simulate the dynamics between the tires and suspension, and prepared the results as an animation in Blender. This is one of the company’s demonstrations of the model working with different tools in a simulation workflow, rather than simply generating visuals.
3D design and browser experiences
OpenAI says Astra and Sol can create interactive browser experiences from a written brief. The announcement features examples including a playable pinball table and a Colosseum that can be viewed from different angles. In the company’s AA-3D evaluation, human evaluators comparing two experiences made from the same prompt gave Astra an Elo score of 1,318. Claude Fable 5.2 scored 1,271, while GPT-6.2 Sol scored 1,259.
Another example cited for Astra is a paper lamp design inspired by the balcony shapes of the Marina City towers in Chicago. In OpenAI’s account, the process runs from reference photos to CAD revisions and a physical lamp assembled by hand. These examples illustrate the models’ claim to turn visual ideas into 3D outputs and presentable interactive experiences; however, the amount of editing required in a production workflow should be assessed for each project.
Impact on architecture and visualization workflows
For architects, interior designers, and visualization teams, the most practically notable part of the announcement is the ability to turn a natural-language brief into an interactive 3D experience and work with components in CAD. This kind of workflow could be explored for early concept presentations, prototypes that communicate design options, or visualizations of mechanical components. The Blender example also shows how modeling, coding, and animation steps can be combined in a single task.
However, the announcement does not provide a compatibility list for specific CAD software, support details for architecture software, hardware requirements, or information about availability in Turkey. Teams should therefore avoid assuming it will work seamlessly in a production pipeline before testing it with their own software and file structures.
Pricing and access
OpenAI says GPT-6.2 Astra’s API input and output token prices are 20% lower than GPT-6 Astra’s, with a 60% discount for cached input. The announced prices are $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. GPT-6.2 Sol’s API token prices remain the same as GPT-6.1 Sol’s. The announcement also points to an option to try Astra in ChatGPT Work, but does not share individual subscription pricing or integration details for architecture software.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
GPT-6.2 Astra’s combination of CAD, simulation, and Blender in a single example workflow is noteworthy, especially for teams preparing proof-of-concept work and presentation prototypes. But for architecture firms, the key measure will be how well it works with their existing software and file structures—not how impressive a demo looks. Since OpenAI has not disclosed compatibility with specific architecture software or hardware requirements, small-scale testing is essential before making purchasing or workflow decisions.
Although API prices have been announced, the total cost depends on how many tokens a task uses; Sol’s API price has also remained unchanged. Access and usage terms for teams in Turkey are not clarified in the announcement either. A more cautious approach would be to test the model on limited, measurable tasks rather than replace existing tools with it.
Frequently asked questions
What can GPT-6.2 Astra do in 3D design?
According to OpenAI, the model can work with components in CAD and simulation tools and create interactive browser experiences from written briefs. In the company’s example, it also animated simulation results in Blender.
How much does GPT-6.2 Astra cost via the API?
OpenAI announced prices of $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. According to the company, Astra’s input and output prices are 20% lower than GPT-6 Astra’s.
Has GPT-6.2 Sol’s API price changed?
OpenAI says GPT-6.2 Sol’s API token prices remain the same as GPT-6.1 Sol’s.



