What's new in OpenAI's GPT-6.2 Astra and Sol models?
What's the story
OpenAI has launched its latest models, GPT-6.2 Astra and GPT-6.2 Sol, for ChatGPT Work and Codex. The new models are designed to complete complex tasks with fewer output tokens, allowing users to accomplish more before reaching their usage limits. This is a major upgrade from the previous versions in the GPT-6 series that were released in September.
Model capabilities
Astra is designed for the toughest projects
The Astra model is recommended by OpenAI for tackling the toughest projects. It can use a screenshot as a template to match layout, colors, and font size.
On the other hand, Sol is the more affordable option for everyday coding and writing tasks. It performed surprisingly well in a coding test against Astra at a fraction of the cost.
Accessibility
New models available on ChatGPT work and Codex
The new models are available on ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
The $200/month Pro plan now includes Ultrafast mode with Sol for the first time. Meanwhile, the $500/month Pro plan gets Ultrafast on both models.
Eligible Enterprise and Edu workspaces can also access this feature once an admin enables it.
Speed boost
Ultrafast mode generates tokens at up to 4.5 times
OpenAI's Ultrafast mode can generate tokens at up to 4.5 times the standard rate on either model, albeit at the cost of higher usage. Developers can access both models through the API as well.
Astra's input and output prices are now 20% lower than GPT-6 Astra, while Sol maintains its GPT-6.1 pricing.
Performance comparison
Astra model aced agents' last exam benchmark
The new Astra model outperformed its predecessor on the Agents's Last Exam benchmark, which tests AI models' ability to handle complex professional tasks.
It scored 5.2 points higher than GPT-6 Astra while costing less than one-third as much to run.
The model also performed better than rival model Claude Fable 5.2 at just 5% of the cost.
Enhanced capabilities
Astra also outperformed GPT-6 Astra on Terminal-Bench-Science 0.1 test
The new Astra model also outperformed GPT-6 Astra on Terminal-Bench-Science 0.1 by 9%. This test measures AI models' ability to carry out scientific research tasks.
OpenAI said that Astra can prepare more detailed and professionally formatted documents, spreadsheets, and presentations for fields such as law, finance, sales, and accounting.
The model also offers improvements in coding, website development, and 3D design.