OpenAI发布GPT-6.2 Astra与Sol,早期反馈不一
CNET报道,OpenAI发布了GPT-6.2 Astra和GPT-6.2 Sol,两款模型在基准测试中超过前代。Astra定位为处理复杂任务,Sol定位为日常使用,目前均面向付费用户,通过ChatGPT Work和Codex提供。早期反馈并不一致。Ethan Mollick表示Astra更难驾驭,而且有时显得“懒惰”;Jason J. Fleagle也批评了模型表现。OpenAI Codex负责人Thibault Sottiaux表示,早期构建版本在发布前进行了更新,并推测此后没有重新测试,因此仍需更多测试才能判断模型的实用表现。
此次发布一方面展现了基准测试提升,另一方面引发了开发者和早期测试者的批评;更新后的模型是否经过重新测试仍不确定,其实际意义尚不清楚。

OpenAI dropped two new AI models this week in the form of GPT-6.2 Astra and GPT-6.2 Sol, each surpassing benchmarks of their predecessors, but people who have begun using them haven’t all walked away very happy.
Like previous models, Astra takes the crown for handling complex tasks, while Sol is considered a more “everyday” model. Currently limited to paid subscribers, the latest models are available in ChatGPT Work and Codex. And while benchmarks may show improvements in a lot of areas, not everyone is seeing what the fuss is all about.
A new AI model seems to drop almost every couple of days from one of the big tech giants, each improving on the previous version in one way or another. However, OpenAI’s latest and greatest are getting some pushback from developers and early testers. For the everyday user, the 6.2 Sol and Astra should perform more than serviceably, but those looking to dig in, test and use the new models for more professional work may run into some hiccups they’re not used to, according to some feedback.
#OpenAI#GPT-6.2 Astra#GPT-6.2 Sol#ChatGPT Work#Codex#Thibault Sottiaux