OpenAI GPT-6.2 Model Fails Early Tests Ahead of Anthropic IPO
OpenAI released its new GPT-6.2 models on Tuesday to the dismay of early testers just before Anthropic prepares for its initial public offering.
Early Reviews Show Major Issues
The new GPT-6.2 Astra and Sol variants were sold as expert assistants for finance and law but received harsh criticism from tech commentators. Matt Shumer stated the models were essentially unusable because they incessantly asked for permission on simple steps like downloading files. He noted the software lacks a concept of autonomy and is leagues behind current Anthropic models in capability and user experience.
Developers Compare Models Directly
McKay Wrigley, CEO of Takeoff, described the release as a major advertisement for Anthropic right before its stock market debut. Ethan Mollick found the space elevator simulation required more guidance than other models to get work done effectively. Shumer conceded that post-training could improve the successor but emphasized the current version feels like a step back from previous OpenAI releases.
Specific Technical Limitations Noted
Company claims that the newer GPT-6.2 Sol manages daily tasks with fewer tokens compared to its earlier version GPT-6.1 Sol because it is faster. Although a negative feedback loop exists, many staff members reportedly set GPT-6.2 Sol as their default choice for handling both technical and non-technical assignments. Yesterday, early reviewers released their assessments following OpenAI's invitation to prominent technology writers to try the fresh model before its public reveal.
Reported by one outlet
Only one outlet has published this. Nothing here has been checked against a second report, so read it as that outlet's account and follow the link below for the original.
Reported by
1 independent outlet. Headline as published. Links open the original report.