‘Google has officially missed its promised launch schedule for its flagship Gemini 3.5 Pro AI model due to severe, ongoing difficulties with code generation. Google CEO Sundar Pichai confidently assured developers in May that the premium tool would roll out in June, but that deadline has passed with zero updates.
A detailed Bloomberg report, citing ten current and former employees, reveals that Google even reset and updated its underlying training data in late June to fix these coding performance issues on Gemini 3.5 Pro. However, the internal results were once again disappointing. The delay has sparked widespread frustration among internal engineers who worry the company is losing its edge to faster rivals.
Internal bureaucracy and corporate resource wars lead to Gemini 3.5 Pro coding issues
The setbacks are heavily tied to Google’s massive, fragmented corporate structure. Currently, separate teams inside Google Cloud, DeepMind, and the Android division are all building their own AI programming tools at the same time.
This siloed approach means competing teams are accidentally duplicating work while aggressively vying for the same limited hardware computing resources. To regain control, Google is currently trying to consolidate these chaotic internal utilities onto a single platform known as Antigravity.
Ironically, this coding roadblock happens while Google’s internal reliance on automation is at an all-time high. Roughly 75% of all new code deployed at the company is now AI-generated and manually approved by engineers, up from 50% last autumn.
However, this shift has caused internal friction. Many software purists inside the company firmly believe that critical systems should remain human-written to maintain strict quality standards.
The competitive pressure
While Google struggles to polish its flagship, leaner AI startups are moving at lightning speed. Anthropic recently deployed its advanced Fable 5 architecture. Conversely, OpenAI released GPT-5.6 Sol, which is optimized for coding and cyber defense. Plus, Chinese lab Moonshot released a massive 2.8-trillion-parameter open-source model called Kimi K3.
Meanwhile, enterprise clients using Google’s lighter Gemini 3.5 Flash model—which rolled out on time in May—are giving mixed reviews. Tech platforms like Figma appreciate its speed-to-quality balance, but the educational platform Platzi explicitly ditched Google for Anthropic’s models. They cited slower performance, structured data errors, and high pricing.
Google has confirmed it is testing the Pro model with corporate partners and the US government. However, its official DeepMind page still simply lists the flagship as “coming soon.”
