Google's Gemini 3.5 Pro Is Months Behind Schedule — and the Coding Gap Is the Core Problem
Bloomberg reports Google's flagship Gemini 3.5 Pro is significantly delayed due to coding performance shortfalls, raising competitive alarms vs. OpenAI and Anthropic.
Update, 17 July 2026: Gemini 3.5 Pro misses its third launch target — what Google has and hasn’t confirmed
Gemini 3.5 Pro did not ship on July 17. This is the third missed date since Sundar Pichai’s “give us until next month” pledge at Google I/O in May, and it extends the central concern of this article: the coding performance gap is not resolved.
The Gemini API changelog carries no entry for a gemini-3.5-pro model ID as of publication. Google’s only public statement remains that the company is “currently testing 3.5 Pro with partners.” No specs, pricing, or benchmarks have been confirmed. Geeky Gadgets reports an August slip is now circulating, though Google has not confirmed a new date.
The competitive window has narrowed further since the original post. GPT-5.6 Sol and Grok 4.5 both launched publicly on July 9, meaning developers are already stabilizing stacks around newer flagships while Google’s Pro tier remains unavailable.
Gemini 3.5 Flash is the only live 3.5-series model. Treat every unconfirmed claim about specs, context windows, or benchmark scores as unverified until Google publishes a model card or API release note.
The original reporting below remains accurate. This update reflects the third delay and the absence of any official launch confirmation.
Google’s most anticipated AI model has a problem, and it is not a small one. Bloomberg reported on July 16 that Gemini 3.5 Pro, the company’s next flagship model, is months behind its original schedule. The headline reason: the model is not good enough at coding, and Google’s attempts to fix that have not yet worked. Alphabet’s stock dropped around 4% on the news.
This matters well beyond the usual “tech company misses a deadline” story. The delay is a signal worth paying attention to if you are building on AI tools, evaluating which models to trust with serious work, or watching the broader race between Google, OpenAI, and Anthropic.
What happened, exactly
Gemini 3.5 Pro was expected to ship in June 2026. At Google I/O in May, CEO Sundar Pichai told developers it was coming “next month.” It did not arrive.
The sticking point is coding performance. According to Bloomberg, sourcing ten current and former Google employees, the model has consistently fallen short of internal targets in this area. In late June, Google updated the training data to try to fix it. The results were disappointing. The company is now reportedly restarting parts of the training process, which is why there is still no release date.
A Google spokesperson did not dispute the delay, saying: “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.”
That is a careful statement. It confirms the model is in testing, not shipped.
Why coding specifically
This is worth slowing down on, because coding is not just one use case among many right now. It is the primary battleground for enterprise AI adoption. Developers, engineering teams, and businesses building AI-powered products are choosing their AI stack based heavily on which model writes the best code.
OpenAI has Codex. Anthropic has Claude Code. Both have established credible positions in this segment. Google, despite its enormous engineering talent and resources, does not have a comparable commercial product. Internal DeepMind staff have reportedly flagged this directly: Google lacks a credible coding product for businesses building AI development tools.
Making things more complicated, Google has multiple teams working on AI coding tools across DeepMind, Google Cloud, and its consumer product divisions, and those teams have not always been working in alignment. Chief AI Architect Koray Kavukcuoglu is now working to unite these efforts, and a dedicated AI coding team within DeepMind has been formed under research engineer Sebastian Borgeaud. But unifying competing internal factions takes time, and time is what Google does not have right now.
The talent picture adds pressure
The delay does not exist in isolation. In the past several weeks, Google has lost some of its most significant AI researchers.
Noam Shazeer, who co-authored the 2017 “Attention Is All You Need” paper that introduced transformer architecture, left for OpenAI. Google had paid $2.7 billion to bring him and part of his Character.ai team back in 2024. He stayed less than two years. John Jumper, who shared the 2024 Nobel Prize in Chemistry for his work on AlphaFold, announced he was leaving for Anthropic. Jonas Adler and Alexander Pritzel, both contributors to AlphaFold research and Google’s AI coding efforts, are also reported to be heading to Anthropic.
These are not peripheral figures. They represent genuine frontier research capability, and their exits have reinforced concerns among remaining staff about Google’s direction and momentum.
What this means if you are building on Google AI today
If you are currently using Gemini 3.5 Flash, the model that is available now, you may already be noticing some of the limitations. Feedback from enterprise users has been mixed. Figma has praised it for balancing speed and quality. But education platform Platzi’s CEO described it as occupying an uncomfortable middle ground: more expensive than the previous 3.1 Flash, slower in practice, and less capable than Anthropic’s Claude 3.5 Sonnet for tasks requiring structured data handling or strong reasoning.
That is a pointed comparison. If you are choosing a model for coding assistance, document analysis, or structured data work, the honest picture right now is that Anthropic and OpenAI are ahead on raw capability. Google’s advantage remains its distribution: Gemini is embedded in Search, Android, and Google Workspace, reaching a vast number of users. Sensor Tower data shows Gemini’s mobile daily active users reached 118 million in June, up 295% year-over-year. But scale of reach and quality of model output are different things, and the gap between them is what this story is really about.
If you are evaluating AI tools for a development team or enterprise workflow, this is not a reason to write Google off. But it is a reason not to wait for Gemini 3.5 Pro before making decisions. OpenAI and Anthropic are shipping now, and their coding-focused products are mature enough to be used in production.
The bigger picture for Google
Alphabet is spending at an extraordinary rate on AI infrastructure, $35.7 billion in data centre and AI investment in Q1 2026 alone, with full-year 2026 capital expenditure guided at $180 to $190 billion. Investors want to see that translate into competitive products. Alphabet’s Q2 earnings are due on July 22, and analysts will be looking closely at Gemini user growth, Google Cloud AI revenue, and any guidance on when 3.5 Pro ships.
The structural challenge Google faces is real. It has more AI surface area than any other company, with Gemini running across Search, Maps, YouTube, Android, and Workspace. Coordinating model releases across that portfolio involves layers of product, legal, safety, and partner review that simply do not exist at a pure-play AI lab. That is not an excuse, but it is context for why the organisation moves differently from Anthropic or OpenAI.
Google has genuine strengths that are not going away: multimodal capability across text, image, and video; deep integration with Google Search data; and serious progress in AI world models. But on the specific question of frontier model capability, particularly for code, the honest answer right now is that its competitors have the edge.
The question is whether Gemini 3.5 Pro, when it eventually ships, is good enough to close that gap, or whether the delay has given OpenAI and Anthropic enough runway to extend it further.