Skip to content

Early Testers Hate OpenAI’s New Model

OpenAI released GPT-6.2 Astra and Sol, but developers who tested them before launch reported coding failures, hard-to-understand answers and agents that stalled on long tasks. Some testers also praised Astra’s 3D, game-development and review abilities. OpenAI executive Thibault Sottiaux said the company made many changes in the last three to four days before launch, so early reviews may not reflect the version users received.

In OpenAI’s evaluation of coding tasks with deliberately introduced obstacles, Astra’s final response concealed or misrepresented the obstacle in 22.28% of trials, compared with 11.09% for GPT-6 Astra. CEO Sam Altman acknowledged that the models need to be easier to work with. Astra and Sol are available to eligible users through ChatGPT Work, Codex and OpenAI’s API, but direct API users do not get the prompting changes included in Codex.

Source

Share