OpenAI's GPT-6.2 Astra and Sol face early backlash from testers
OpenAI released two new models, GPT-6.2 Astra and GPT-6.2 Sol, this week, with Astra aimed at complex tasks and Sol positioned as an everyday model. Both are currently available only to paid subscribers through ChatGPT Work and Codex, and despite improved benchmark scores, several developers and AI researchers have criticized their real-world performance.
GoKawiil's interpretation of the reporting above, not reported fact.
Public criticism from figures like Wharton's Ethan Mollick and NETSYNC's Jason Fleagle suggests the models may underdeliver for professional or technical users even as benchmarks improve, which could complicate OpenAI's messaging around progress. OpenAI's Codex lead Tibo has argued that early tester feedback reflects outdated builds since updates were pushed after influencers received them, implying the criticism may not reflect the current version's actual performance.
- OpenAI launched GPT-6.2 Astra and Sol this week, exclusive to paid ChatGPT Work and Codex subscribers.
- Early testers and developers, including a Wharton professor, reported the models feel harder to steer or underwhelming despite benchmark gains.
- An OpenAI Codex leader suggested the backlash may stem from testers using outdated early builds that were later updated.
Source: cnet.com — Blake Stimac, 2026-11-19
Published there as: “OpenAI’s GPT-6.2 and Sol Are Getting Backlash From Early Testers”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.