Gemini 3.5 Pro slips months behind as coding falls short of internal goals

CNBC and the LA Times report that Gemini 3.5 Pro — expected to feature a 2-million-token context window (double its predecessor) and a new 'Deep Think' multi-step reasoning layer — has slipped months as engineers struggle to hit internal coding targets. Alphabet shares dropped on the news, reflecting investor anxiety that Google is falling behind OpenAI and Meta at the exact moment open-weights models like Kimi K3 are compressing the field.

The LA Times' inside account describes clashing teams and frustrated engineers, with Chief AI Architect Koray Kavukcuoglu working to unify the company's fragmented internal AI coding tools. The model was reportedly delayed specifically to fix structural issues rather than ship something that underperforms on the agentic-coding workloads that now define competitiveness.

Separately, Google rebranded NotebookLM as Gemini Notebook, with Gemini 3.5 and 'Antigravity' upgrades slated for AI Pro subscribers, and Logan Kilpatrick announced new cost controls, a free tier, and scheduled triggers for managed agents in the Gemini API — signs the product surface keeps moving even as the flagship model stalls.

Competitively, the delay is damaging optics: coding is the battleground where Anthropic's Claude and OpenAI's Codex have set the pace, and a public stumble here feeds the narrative that Google's research strength isn't translating into shippable frontier models. Sundar Pichai touted Intel's use of Gemini Enterprise for semiconductor design this week, but enterprise adoption anecdotes can't fully offset a flagship miss. Watch whether Google ships a scoped 3.5 Pro in the coming weeks or waits — every week of delay hands rivals a cleaner story.