Google Delays Gemini 3.5 Pro Over Coding Issues as Internal Frustration Mounts

Google's flagship AI model Gemini 3.5 Pro remains unreleased months after the company promised a June 2026 launch, citing coding performance problems that have yet to be resolved.

The delay is more than a missed deadline. It reflects deepening concerns inside Google about whether the company is keeping pace with rivals Anthropic and OpenAI in one of the most competitive technology races of the decade. For businesses and developers building on Google's AI infrastructure, the wait signals genuine uncertainty about when the company's most capable model will arrive — and what it will deliver when it does.

---

What Google Promised and What Happened

Google announced the Gemini 3.5 series at its annual I/O developer conference and released Gemini 3.5 Flash the same day. In a blog post dated May 19, 2026, the company stated that Gemini 3.5 Pro was already in internal use and that Google looked "forward to rolling it out next month."

June came and went without a release. As of publication, the Gemini API release notes carry no entry for a Gemini 3.5 Pro model, and Google has not announced a replacement shipping date.

A Google spokesperson told Bloomberg the company is "currently testing 3.5 Pro" with partners, along with an upgraded Flash model. Testing with partners is not a public rollout, and the statement offered no timeline for when that changes.

Until Pro ships, Gemini 3.5 Flash remains the only model Google has released in the 3.5 series. At I/O, Google made 3.5 Flash the default model in AI Mode globally, so the delay does not immediately affect Search results. What it does affect is developer and enterprise planning built around Google's stated roadmap — a distinction that carries significant weight for organisations that have already committed resources based on that guidance.

---

Coding Performance Is at the Centre of the Problem

Bloomberg reporters Julia Love and Davey Alba broke the delay story, citing people familiar with the matter. Ten current and former Google employees described frustration inside the company. The people declined to be named when discussing internal concerns.

One person told Bloomberg that Google updated the training data for Gemini late last month specifically to improve coding skills — and that the results were disappointing. Coding performance is now identified as one factor holding up the 3.5 Pro release.

This is not the first time Google has acknowledged a gap in coding capability. In May, Google CEO Sundar Pichai stated publicly that the company was "a bit behind" the frontier on agentic coding. Pichai tied that gap to the fact that Google lacks the kind of developer-facing coding product that generates the training data competitors benefit from. Bloomberg is now reporting that broader coding performance remains unresolved enough to delay the flagship model entirely.

Search Engine Journal's June coverage of two senior departures from Google's AI organisation also cited Bloomberg reporting on concerns inside Google DeepMind about what the company offers businesses building AI coding tools. The departures and the delay together suggest the internal pressure is real and sustained. For organisations already navigating the risks and challenges of deploying AI in business, this kind of instability in a core vendor's roadmap adds a further layer of complexity to planning.

Why Coding Capability Matters So Much

Coding performance has become a primary benchmark by which enterprise buyers and developers evaluate large language models. The shift toward agentic AI workflows — where models autonomously write, test, and debug code — means that a gap here is not a marginal limitation. It affects the core use case driving AI adoption across engineering teams and software businesses alike.

Google's acknowledged shortfall in this area is particularly consequential because agentic coding is where the competitive battle is most intense right now. Anthropic's Claude and OpenAI's models have been iterated rapidly with developer feedback loops that Google, without a comparable coding product, has struggled to replicate. That structural disadvantage is now manifesting directly in the delay of its most important near-term model release. You can see how this plays out in practice by looking at how businesses are currently using artificial intelligence — coding assistance and agentic development tools feature prominently among the highest-value applications.

---

What the Delay Means for Developers and Businesses

The situation has a familiar shape for anyone who has followed product launches in the technology industry: a model confirmed to exist in testing, a launch window that has closed, and no revised date on the calendar. The product is real. The gap between internal readiness and public availability is simply wider than Google indicated it would be.

For teams building on Google's AI stack, the practical impact is direct. Any roadmap that assumed Gemini 3.5 Pro would be available by mid-2026 needs to be revisited. Google has not named a new month, and the Bloomberg report does not offer one either. Anything circulating about an arrival date traces back to third-party reporting rather than to Google.

The Competitive Pressure Is Real

The competitive context sharpens the concern. Anthropic and OpenAI have continued shipping models during the period Google has spent working through its coding challenges. Employees cited in the Bloomberg report described frustration that Google is losing ground as rivals release models that outperform Gemini in key benchmarks.

For businesses evaluating AI platforms, the delay is a factor worth weighing carefully. Google remains a major player with significant infrastructure advantages, but the gap between announced timelines and actual delivery is now measurable in months. Organisations considering a long-term commitment to any single AI platform should account for the barriers to AI adoption that vendor delays and roadmap uncertainty introduce — particularly when the pace of the broader market means that a months-long slip can meaningfully affect what is possible today versus what was planned.

How Different Audiences Should Respond

• Developers and engineering teams should pause any product planning contingent on Gemini 3.5 Pro availability and build flexibility into timelines until Google announces a confirmed release date.
• Businesses evaluating AI vendors can use this delay as a prompt to benchmark current Gemini 3.5 Flash performance against Anthropic and OpenAI offerings — making platform decisions based on what is available now rather than what has been promised.
• Marketing and SEO professionals relying on AI Mode in Google Search can note that 3.5 Flash remains the active model and that no change to Search-generated answers is expected as a direct result of the Pro delay.

For broader context on how this competitive dynamic is reshaping enterprise AI decisions, Google's own AI overview documentation provides useful background on how the company frames its development priorities and responsible release approach.

---

Looking Ahead

Google's position heading into the second half of 2026 is that its most capable Gemini model is in partner testing with no public timeline attached. The company has not clarified whether the coding issues identified in late June training updates have been addressed or are still being worked through.

The Gemini 3.5 Flash model continues to serve as the default in AI Mode and remains available through the Gemini API. For most Search users, nothing changes today. For developers and enterprises waiting on Pro, the situation remains open-ended — and the absence of a revised commitment date from Google means the uncertainty is unlikely to resolve quickly.

What this moment underlines is the importance of building AI strategies around demonstrated capability rather than announced roadmaps. Vendor timelines shift. Competitive pressures create unpredictable delays. The organisations best positioned to navigate that environment are those evaluating what models can do today, maintaining optionality across providers, and treating promised release dates as signals rather than certainties.