Google's Gemini 3.5 Pro Delay Exposes the Limits of Scale

Google’s next major AI model, Gemini 3.5 Pro, is months behind schedule. According to Bloomberg, which cited people familiar with the matter, the delay stems from persistent shortfalls in coding performance—an area that has become one of the most important benchmarks in the industry.

The model was originally expected to arrive in June, shortly after Google announced the 3.5 series at its I/O conference. Only the lighter Flash variant has shipped. In late June, Google tried to close the gap by updating the model’s training data with a focus on coding skills. The results disappointed internal teams. The company is now testing 3.5 Pro with select partners while also working on an improved Flash model, but it has not set a new public release date.

The setback has generated frustration inside Google. Multiple current and former employees told Bloomberg they worry the company is losing ground as Anthropic and OpenAI continue to release models that perform better, particularly on coding and agentic tasks. Leadership has previously acknowledged that Google was “a bit behind” on agentic coding, in part because it lacked the tight product feedback loops that competitors enjoy from large developer user bases.

Why Scale Has Not Delivered

Google possesses advantages that should be decisive: the world’s largest search index, YouTube, Android, Maps, vast amounts of proprietary data, enormous computing resources, deep research talent, and essentially unlimited capital. Yet those advantages have not produced the fastest or most capable models in the areas that matter most right now.

Coding ability and agentic performance have become central to how enterprises and developers evaluate frontier models. Models that can reliably write, debug, and execute code are winning mindshare and contracts. In this environment, speed of iteration and focused product loops appear to outweigh raw scale.

OpenAI and Anthropic have repeatedly demonstrated that smaller, more coherent organizations can move faster. Both companies treat coding performance as a core priority and ship improvements at a cadence that forces competitors to react. xAI’s Grok has followed a similar path, emphasizing capability and rapid releases over the complex integration challenges that come with operating a global search and consumer technology company.

Google’s structure works against it in this race. Multiple groups—DeepMind, Google Cloud, Android, and others—have pursued overlapping AI and coding initiatives. Aligning priorities across those organizations slows decision-making and dilutes focus. The result is a pattern of announcements followed by delayed flagship models that fail to match the practical performance of rivals.

The Cost of Falling Behind

The delay is not merely a scheduling issue. It risks further erosion of developer preference at a moment when coding and agentic workflows are becoming foundational to how AI is used in production. Once developers and enterprises standardize on a particular model family for these tasks, switching costs rise. Google’s historical strength in distribution and data may matter less if the models themselves lag on the capabilities people actually use day to day.

None of this means Google cannot recover. Its research depth and infrastructure remain formidable. But the current trajectory suggests that resources alone are not enough. In the present phase of the AI race, organizational focus and shipping velocity have proven more important than the size of a company’s data centers or the breadth of its data assets.

Until Gemini 3.5 Pro actually arrives and demonstrates clear leadership on coding and agentic benchmarks, the market will continue to favor the companies that have moved with greater urgency and clearer priorities.