en.Wedoany.com Reported - Google is restructuring its AI coding team, formed just a few months ago, to close the gap with Anthropic in the most commercially valuable application today—code writing. According to The Information, this temporary "task force" is expanding into a longer-term division focused on the "intermediate training" phase, the model training stage between pre-training and post-training.

This adjustment aims to change how Gemini models are trained. The new team's goal is not just to optimize the ready-made answers the model provides, but to embed core capabilities earlier: strengthening coding skills and extending these abilities to related business tasks, even including creating presentations. This aligns with the path already taken by Anthropic and OpenAI, where coding tools go beyond programming itself.

Google essentially acknowledges that in this niche, a powerful foundation model alone is not enough to win; it requires specialized training, dedicated computing resources, and tighter product iteration cycles. This restructuring comes against an unsettling backdrop: the company recently lost two prominent researchers. Noam Shazir joined OpenAI, reportedly because the computing power needed for his project was reassigned to other teams; Nobel laureate and AlphaFold lead John Jumper is about to join Anthropic, having been transferred to the coding team shortly before his departure.

Gemini model training has not been smooth sailing, with indirect signs reflected in the progress of its flagship product. The Gemini 3.5 Pro, announced at the I/O conference in May, has yet to be publicly released. According to Business Insider, its launch has been delayed to July. Google is collecting feedback on its Antigravity and LMArena platforms, further training the model to handle longer agentic tasks, and separately addressing token consumption issues. Currently, only Gemini 3.5 Flash is publicly available, while the Pro version remains limited to enterprise previews.

Google's safety margin remains substantial, with its self-developed TPU chips, Gemini models, cloud services, search, YouTube, and a vast distribution network. The question is whether this training restructuring will translate into meaningful quality improvements, or just a change of signage. The first test will be the release of Gemini 3.5 Pro in July.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com