Google has delayed the broader rollout of Gemini 3.5 Pro, its next flagship model, after internal testing found it falling short of the company's own bar on coding and complex, long-horizon reasoning, Bloomberg reported Thursday. The model remains in limited enterprise preview while engineers keep working on it, and Google has not announced a new target date.

A missed timeline

Google unveiled Gemini 3.5 Flash at I/O 2026 in May and said at the time that the Pro version would follow in June. That date came and went. According to Bloomberg's reporting, Google updated the data used to train Gemini in late June specifically to improve coding performance, but the results of that effort disappointed internal testers. Gemini 3.5 Pro was also expected to ship with a "Deep Think" reasoning mode aimed at complex scientific, mathematical and coding problems; people familiar with testing said its performance on multi-step, long-horizon tasks fell short of the bar Google had set for the I/O release.

A Google spokesperson did not confirm a delay outright but told reporters: "We're shipping quickly across a wide range of models while keeping them highly cost-effective for customers. We're currently testing 3.5 Pro, an upgraded Flash model, and other models with partners."

Why coding is the sticking point

The gap centers on agentic coding — the kind of multi-step, tool-using code generation that has become the main battleground between frontier labs. At I/O in May, CEO Sundar Pichai acknowledged Google was "a bit behind" the frontier on agentic coding, a shortfall he tied partly to Google lacking a developer-facing coding product on the scale of rivals' tools, which generate the kind of real-world usage data that helps improve those capabilities.

Why it matters

The delay lands in the middle of an unusually crowded stretch of frontier releases: xAI shipped Grok 4.5, OpenAI took its GPT-5.6 family public, and Moonshot AI released the open-weight Kimi K3, all within about two weeks of each other in early-to-mid July. A slip on coding and reasoning benchmarks — the two areas enterprise and developer customers scrutinize most when picking a model for agentic workflows — is a notable stumble for Google at a moment when competitors are moving fast on exactly those fronts. Google has given no indication of when Gemini 3.5 Pro will see a full public release.