AI & Tech · Paris
OpenAI releases GPT-6.2 Astra and Sol, claiming a lead over Claude Fable 5.2
Both models are open to paid ChatGPT Work and Codex users and to developers through the API, with Astra aimed at complex projects and Sol at everyday tasks. The launch comes nearly two months after OpenAI dropped an earlier Astra update to add stronger safeguards.
AI text based on reports by TechRadar, EL PAÍS, Bloomberg Business and other publishers. Published . How it is made
For membersThe first 30 seconds for everyone. Synthetic voice (AI).
In brief
From 17 November 2026, Astra and Sol are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API, while Free accounts are excluded.
On Agents' Last Exam, OpenAI says Astra leads GPT-6 Astra by 5.2 points and Claude Fable 5.2 by 5.5 points, at 5% of that rival's cost per task.
The $200 per month Pro plan gets Ultrafast mode with Sol for the first time, which TechRadar says runs up to eight times faster than GPT-6 Astra in Standard mode in Codex.
- September 2026GPT-6 generation begins with Astra
- 17 NovGPT-6.2 Astra and Sol released to paid users
Two models for paid subscribers
OpenAI has released GPT-6.2 Astra, built for complex projects, and GPT-6.2 Sol, aimed at everyday work such as analysis, writing and coding. Both became available on 17 November 2026 in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and developers can reach them through the API. Free accounts cannot use either model, and access is limited to Work mode rather than standard chat. The GPT-6 line began with Astra in September, and Sol moved to GPT-6.1 a few weeks later at DevDay. OpenAI says both models deliver better results on demanding tasks while spending fewer output tokens, which leaves users more room under their usage limits.
On the product side, the company says the new Astra produces slides, spreadsheets and documents that look less machine-made and follow templates more closely. It also says Astra handles website, game and 3D design code better than its predecessor. For coding work on existing projects, OpenAI reports that Astra is stronger at making and testing changes, and it credits early testers with the result.
“Early testers reported cleaner code style and more focused tests, with fewer rounds of revision,”
Benchmarks and pricing
OpenAI reports that on Agents' Last Exam, Astra scored 5.2 points above GPT-6 Astra at less than a third of the estimated API cost per task. The same test places Astra 5.5 points ahead of Claude Fable 5.2, at a cost of 5% of that rival's. El País reports that OpenAI also claims Astra is the strongest model for scientific research, citing new records on Terminal-Bench Science 0.1 and the Vals Science Index. On FrontierCode 2.0, the coding evaluation from Cognition, Sol came within 0.5 points of Astra at roughly a fifth of the estimated cost per task, according to OpenAI. Astra's standard API input and output prices are 20% lower than those of GPT-6 Astra, while Sol keeps the GPT-6.1 pricing.
- vs GPT-6 Astra
- 5.2 points
- vs Claude Fable 5.2
- 5.5 points
Data from the article text
Ultrafast access widens
Subscribers to the $200 per month Pro plan receive Ultrafast mode for the first time, and it works with Sol. Until now, Ultrafast was limited to the $500 per month Pro tier. TechRadar reports that Ultrafast handles requests up to eight times faster than GPT-6 Astra in Standard mode in Codex. A separate figure from OpenAI puts token generation at up to 4.5 times the standard token rate on either model, with the trade-off that it draws down more of a user's allowance. Eligible Enterprise and Edu users gain access once a workspace admin switches it on, and speeds can be changed from the model picker.
Paris rebuilt for browser exploration
OpenAI developer Thomas Ricouard asked GPT-6.2 to recreate Paris at full scale so he could walk or drive through it in a browser. The model combined street maps with French survey data covering more than 131,000 buildings. Ricouard then took screenshots to show the model which changes he wanted. The result, called Paris Atlas, takes a while to load. One reviewer who tried it described choosing a spot on the map, being placed on a realistic street and walking with the arrow keys while navigating around 3D pedestrians.
Rivalry with Anthropic
Bloomberg reports that the release came nearly two months after OpenAI scrapped a previous Astra update to establish stronger safeguards for the technology. El País frames the launch as a direct challenge to Anthropic, which is in the middle of a stock market flotation, and notes that OpenAI's comparison targets Claude Fable 5.2 specifically. Bloomberg describes Astra as OpenAI's most capable model for complex professional projects. Sam Altman leads OpenAI, and the company's comparative claims remain its own figures, with no independent testing cited in the articles.
Ask about this text
Sources and context
- TechRadarOpenAI just launched GPT-6.2, and it's powerful enough to recreate the whole of Paris as a virtual world you can explore in your browser
- EL PAÍSOpenAI pressures Anthropic during its stock market launch with its new version of AstraOpenAI presiona a Anthropic en plena salida a Bolsa con su nueva versión de Astra · original in Spanish
- Bloomberg BusinessOpenAI Unveils New Astra AI Model After Scrapping Prior Release
- 9to5MacOpenAI upgrades ChatGPT and Codex with new GPT-6.2 Astra and Sol models
- Sam Altman
- Leads OpenAI, the company behind the GPT-6.2 Astra and Sol models
- Places
- Paris
Pollar is free and carries no ads in the text. Its members keep it going.
Support Pollar