Google Renames NotebookLM and Debuts Gemini 3.5 Pro [Model Behavior]
Welcome to Model Behavior. I am Nina Park. Today we examine how AI systems are built and deployed in professional environments. On this Friday, July 17th, 2026, Google has initiated a major realignment of its product suite. This includes a rebranding of its most successful research tool and the release of a long-awaited frontier model that had been delayed by technical hurdles. <br/><i>acting_description:</i> professional, steady, leading <i>speed:</i> 0.98 <i>trailing_silence:</i> 0.4 I am Thatcher Collins. It has been a busy morning for the teams at Google DeepMind and Labs. We are looking at two major developments: the transition of NotebookLM into Gemini Notebook, and the arrival of Gemini three.five Pro. This rollout is significant because Google is playing catch-up on a release schedule that slipped by six weeks over the last month. <br/><i>acting_description:</i> engaged, grounded, responsive <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 Let us start with the product evolution. Google is renaming NotebookLM to Gemini Notebook. This tool started as Project Tailwind back at I/O 2023 and has grown to 30 million users and 600,000 organizations. According to Google, while it remains a standalone product, the integration into the broader Gemini ecosystem is becoming explicit, moving away from its experimental roots into a core productivity tool. <br/><i>acting_description:</i> confident, clear, measured <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.5 The technical shift is more than just a name change, Nina. Google is introducing a secure cloud computer for every notebook. This allows Gemini Notebook to natively write and execute code. For researchers, this means you can perform complex data analysis that is grounded specifically in the sources you have uploaded, rather than relying on the general knowledge base of the model. <br/><i>acting_description:</i> questioning, sharp, measured <i>speed:</i> 0.97 <i>trailing_silence:</i> 0.4 That is a major shift for a tool once seen as a simple summary engine. It is moving into the realm of active computation. However, these features are not universal yet. The integrated code execution is available today for AI Ultra and Workspace business users, while the rollout to standard Pro users on the web is scheduled for the coming weeks. <br/><i>acting_description:</i> authoritative, steady, professional <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 It is a clear monetization play. But the bigger technical story today is Gemini three.five Pro. It officially launched this morning, July 17th. This is the model promised back in June, and as reported, it arrives roughly six weeks late. It is landing in a market where GPT-five.six Sol and Claude Fable five have already been setting the pace for several weeks. <br/><i>acting_description:</i> grounded, sharp, inquisitive <i>speed:</i> 0.98 <i>trailing_silence:</i> 0.4 The standout specification for three.five Pro is the context window. We are looking at two million tokens. To put that in perspective, Thatcher, that is ten times the capacity of Claude Fable five. Google is positioning this as the premier tool for professionals who need to process entire codebases or massive research libraries within a single prompt. <br/><i>acting_description:</i> clear, leading, confident <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.5 The context window is impressive, but we have to look at the reasoning layer Google calls Deep Think. This is their answer to the complex, multi-step problem solving seen in recent competitor releases. But here is the friction: Deep Think is gated behind the Ultra subscription tier, which costs 250 dollars per month. That is a high barrier to entry compared to other models. <br/><i>acting_description:</i> responsive, measured, questioning <i>speed:</i> 0.96 <i>trailing_silence:</i> 0.4 It is also worth noting why this took so long. Reports indicate DeepMind had to scrap the original Gemini two.five Pro architecture entirely. They encountered structural failures in recursive tool-calling and SVG scene generation. They basically had to rebuild the foundation mid-cycle because the previous iteration was not meeting the frontier performance standards set by their peers earlier this summer. <br/><i>acting_description:</i> steady, professional, measured <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 That explains the delay, but it puts Google in a tough spot. They are the fourth major lab to ship a frontier model this season. While the two million token window is a unique selling point, the market has already had time to stabilize around other ecosystems. Google is asking users to switch to a more expensive tier for a model without a track record. <br/><i>acting_description:</i> sharp, engaged, grounded <i>speed:</i> 0.99 <i>trailing_silence:</i> 0.4 It is a high-stakes move to see if sheer context size can outweigh a late arrival and premium pricing. For professionals who need t