Google has quietly missed its self-imposed June deadline for Gemini 3.5 Pro, the next-generation AI model it positioned as a major leap forward at I/O on May 19. According to Bloomberg, the holdup comes down to disappointing coding performance—a problem that even a late-June retraining of the model’s underlying data couldn’t fix. No new launch window has been given, leaving developers, IT administrators, and Windows users who build on Google’s AI stack with a clear message: Gemini 3.5 Pro isn’t coming anytime soon.

The Silent Deadline Slip

Google’s announcement at I/O 2026 was unambiguous: Gemini 3.5 Flash would ship immediately, and the more capable Pro variant would follow “next month.” Flash did launch—it’s now live across the Gemini app, Google Search’s AI Mode, AI Studio, Android Studio, Antigravity, and enterprise products. But as of July 19, Gemini 3.5 Pro is nowhere to be found in the official DeepMind model catalog. Only Flash appears there, confirming that the model hasn’t reached broad availability for consumers, developers, or Google Cloud customers.

The delay isn’t just a missed date. Bloomberg, as summarized by 9to5Google and Android Authority, reports that Google specifically retrained or updated Gemini’s training data in late June in an effort to boost its coding capabilities. The results fell short of internal expectations. Google hasn’t confirmed those technical details, but it did tell 9to5Google that it is “currently testing 3.5 Pro, an upgraded Flash model, and other models with partners.” That testing phase, typically a precursor to public release, appears to have been extended indefinitely.

Why Coding Performance Matters

Coding wasn’t supposed to be the stumbling block. Google marketed Gemini 3.5 Flash as its “strongest coding and agentic model” yet, tuned for long-running automated workflows, codebase maintenance, and multi-step development tasks. Flash was supposed to be the lighter, faster sibling; the Pro version was expected to deliver even better reasoning and code generation.

Yet Bloomberg’s reporting indicates it’s precisely those high-end coding skills that aren’t meeting the bar. For a model that Google initially said was already in internal use—implying it was working well enough for Google’s own engineers—the public delay suggests a gap between internal expectations and shipped reality. That gap matters for anyone who uses AI to generate, review, or deploy code. A model that stumbles on complex logic, multi-file refactors, or domain-specific languages can introduce subtle bugs that are hard to catch.

The timing also comes amid a broader internal push at Google to consolidate its AI coding tools. Multiple teams—DeepMind (AI Studio), Cloud (Vertex AI), and Android (Android Studio)—have been building parallel tooling, sometimes competing for compute resources. Google is reportedly steering efforts toward Antigravity, its agent-oriented development platform, but that consolidation won’t solve the immediate quality problem with Pro.

Impact on Developers and IT Administrators

For Windows users who write code, manage dev teams, or integrate AI into enterprise workflows, the practical fallout is straightforward. Gemini 3.5 Pro isn’t an option right now, and there’s no reliable delivery date. The available lineup remains:

  • Gemini 3.5 Flash: Available now, marketed for speed and agentic tasks.
  • Gemini 3.1 Pro: The previous flagship, dating back to February 2026.
  • Competing models: OpenAI’s GPT-4o, Anthropic’s Claude, and GitHub Copilot’s underlying models all offer strong coding performance and have shipped updates recently.

Developers shouldn’t plan new editor integrations, CI/CD pipelines, or internal tooling around Gemini 3.5 Pro until Google publishes availability and pricing. IT administrators evaluating enterprise AI subscriptions should treat the current Gemini portfolio as static for the time being. That doesn’t mean you should ignore Gemini entirely—Flash may still be a good fit for lightweight coding tasks or agent workflows—but it does mean benchmarking against real repositories is essential before committing.

Home users who interact with Gemini via the web app or mobile won’t notice any difference; Flash already powers those experiences. The delay primarily affects power users, developers, and orgs that need the higher reasoning ceiling Pro was supposed to deliver.

How Did We Get Here?

Google’s push to infuse code generation into its AI models has been rapid. By April 2026, the company reported that 75% of all new code at Google was AI-generated and reviewed by engineers—up from just 50% in the fall of 2025. That internal reliance suggests Google itself depends on the very capabilities that now appear to be holding back the public release.

The missed June deadline isn’t Google’s first model delay, but it’s notable because the company explicitly tied the launch to a specific window. At I/O, executives said Pro was “showing great improvements.” Retail confidence like that makes a silent miss more jarring.

Behind the scenes, there’s also tension: some Google engineers hold a more traditional view that all important code should be human-written, according to ex-employees cited by Bloomberg. That philosophical split may have contributed to a higher internal quality bar—one that the current Pro model can’t yet clear.

What to Do Right Now

If you’re a Windows developer or IT decision-maker, here’s a concrete checklist:

  • Pin production integrations to currently available models. If you’re using Gemini via API, Vertex AI, or AI Studio, keep your apps pointed at gemini-3.5-flash or gemini-3.1-pro. Don’t write code that depends on a Pro model ID that doesn’t exist yet.
  • Test Flash against real workloads, not benchmarks. Google’s own benchmarks show strong results, but independent testing on your own repositories—especially tasks involving complex logic, error handling, or cross-file changes—will reveal whether Flash is sufficient.
  • Avoid budgeting around Pro. Until there’s a firm release date and pricing, assume the cost and capability profile you have today will persist. Don’t factor a “Pro upgrade” into project timelines or cost models.
  • Maintain fallback providers. If your workflows involve automated code generation or review, keep an alternative model in place (e.g., GPT-4o, Claude) so you’re not disrupted if Flash doesn’t meet your quality bar.
  • Watch for partner testing news. Google confirmed it’s testing with partners; if you’re a cloud or Workspace enterprise customer, ask your account team for non-public guidance on timing.

What Comes Next

Google hasn’t cancelled Gemini 3.5 Pro. The company’s statement emphasizes it is still testing and shipping multiple models “quickly” while keeping them cost-effective. But the lack of a revised deadline suggests the coding issues are deeper than a simple tuning tweak. A retrained model that still disappointed in June points to systemic challenges—possibly in training data quality, model architecture, or the difficulty of exceeding the current state-of-the-art in code generation.

There’s also the matter of internal consolidation. Should Google succeed in merging its disparate coding tools under Antigravity, future launches might be more coherent. But that’s a longer-term play. In the near term, the AI coding race won’t wait. Competitors are iterating rapidly, and developers who rely on cutting-edge AI will gravitate toward whichever platform ships capable, reliable models first.

For now, the advice is patience—but not passivity. Test what’s available, prepare to switch if needed, and ignore the hype until the code itself proves the model’s worth.