Google's Gemini 3.5 Pro Slips Months Behind Schedule Over Coding and Reasoning Gaps

Google’s flagship AI model isn’t ready, and the company has stopped pretending otherwise. On July 16, 2026, Bloomberg reported that Alphabet has pushed Gemini 3.5 Pro’s release back months past its original timeline, after internal testing showed the model falling short on coding tasks and on complex, multi-step reasoning work. Alphabet shares fell about 4% that day, erasing roughly $200 billion in market value in a single session.

Sundar Pichai told developers at Google I/O in May that Gemini 3.5 Pro would ship “the following month.” June came and went with no release. A second target, leaked as July 17, also passed. By late July, Google had shipped three new Flash-tier models, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a security-tuned Gemini 3.5 Flash Cyber, while the actual flagship stayed in partner testing with no public date attached.

TL;DR

Google’s most capable model, Gemini 3.5 Pro, is running months late. Bloomberg’s Julia Love and Davey Alba broke the story on July 16, citing ten current and former Google employees, and reported that a late-June effort to retrain the model on new coding data produced results one source called “disappointing.” Pichai has already conceded Google is “a bit behind” on agentic coding relative to rivals. The timing is rough: OpenAI shipped GPT-5.6 publicly on July 9, and Anthropic followed with Claude Opus 5 on July 23-24, both landing while Gemini 3.5 Pro sat untested outside partner channels. Google’s public line is that it’s still testing the model and will ship “when it’s ready.” Three missed deadlines in, that phrase is carrying a lot of weight.

What was reported

• The source: Bloomberg’s Julia Love and Davey Alba broke the story on July 16, 2026, citing ten current and former Google employees who described internal frustration with the timeline.
• The core problem: per their reporting, coding performance and complex reasoning work fell short of what Google expected internally, not any single catastrophic failure.
• The fix that didn’t fix it: in late June, Google updated the training data feeding Gemini in an attempt to close the coding gap. One source called the results disappointing.
• Two missed targets: Pichai promised a June rollout at I/O in May. A second, leaked target of July 17 also came and went without a launch.
• Flash ships instead: on July 21, Google released three new models, none of them Gemini 3.5 Pro. In the same announcement, Google said it has begun “our most ambitious pre-training run yet,” referring to work on Gemini 4, a line that suggests some internal attention has already moved past 3.5 Pro.
• The official statement: a Google spokesperson told Bloomberg the company is “currently testing 3.5 Pro, an upgraded Flash model, and other models with partners,” and is “shipping quickly across a wide range of models while keeping them highly cost-effective for customers.”
• No new date: as of this writing, Google has not given a revised public launch window for Gemini 3.5 Pro.

The stock market did not shrug this off. Alphabet fell about 4% on the July 16 report, a one-day drop larger than the company’s entire full-year capital spending guidance of $180 to $190 billion.

The talent backdrop

This delay didn’t happen in a vacuum. In June, a cluster of senior DeepMind researchers left for competitors within roughly a week of each other. AlphaFold lead John Jumper departed for Anthropic after nearly nine years at the lab. Fellow AlphaFold contributors Jonas Adler and Alexander Pritzel followed him there. Separately, Gemini co-lead Noam Shazeer left for OpenAI. Fortune reported Alphabet shares fell 5-6% around June 22 as those departures became public, with coverage tying the drop to concerns about the company’s ability to retain senior AI talent.

Two stock hits inside a month. Both about whether Google can keep its best people and its best model on schedule.

Why it matters

Coding has become the benchmark that matters most in the current model cycle. Enterprise buyers evaluate frontier models on how well they handle agentic coding workflows, and both Anthropic and OpenAI have leaned into that as their primary pitch through 2026. Pichai himself has acknowledged the gap. When the CEO says his own company is “a bit behind” on the metric everyone is racing on, a delay stops being a scheduling footnote and starts looking like a quality-bar problem.

The competitive calendar makes it worse. OpenAI’s GPT-5.6 rolled out publicly on July 9. Anthropic’s Claude Opus 5 launched July 23-24, priced at $5 per million input tokens and $25 per million output, with a 1-million-token context window. Both shipped inside the same three-week window that Gemini 3.5 Pro was supposed to occupy. Google isn’t just late. It’s late while its two biggest rivals both cleared the bar it’s still trying to hit.

There’s a reading of this that’s actually reassuring, though. Google scrapping a training run and restarting rather than shipping a model that underperforms its predecessor is the kind of discipline every lab claims to have and few demonstrate under pressure. A rushed Gemini 3.5 Pro that flopped on independent coding benchmarks would have been a worse story than a delayed one. Whether the market agrees is a separate question, and for now the market clearly does not.

For AI search specifically, the near-term impact is limited. Gemini 3.5 Flash, not Pro, is the model currently running Google’s AI Mode and AI Overviews, and that hasn’t changed. What the delay does mean is that whatever reasoning uplift Pro was supposed to bring to Google’s most complex, multi-step search answers is now pushed out along with everything else. Sites optimizing for AI Overviews citations aren’t dealing with a new model behavior yet. They’re dealing with the current one, for longer than Google planned.

Frequently asked questions

Why was Gemini 3.5 Pro delayed? Bloomberg reported that internal testing showed the model falling short of Google’s expectations on coding tasks and complex reasoning work. A late-June retraining effort aimed at fixing the coding gap produced results one internal source described as disappointing.

When will Gemini 3.5 Pro actually launch? Google has not given a public date. The company says it’s testing the model with partners alongside other releases. Two previous targets, a June rollout promised at I/O and a leaked July 17 date, both passed without a launch.

Does this affect Google Search or AI Overviews right now? Not directly. Gemini 3.5 Flash remains the model running AI Mode and AI Overviews, and that has not changed because of the Pro delay. The effect is more about a missed upgrade than a current outage or regression.

How does this compare to GPT-5.6 and Claude Opus 5? Both shipped in the same window Gemini 3.5 Pro was supposed to occupy. OpenAI’s GPT-5.6 rolled out publicly July 9, and Anthropic’s Claude Opus 5 launched July 23-24. Google’s flagship remains in partner testing while both competitors are live.

Has Google delayed a flagship Gemini model before? Google has shipped incremental Gemini updates on a fairly tight cadence through 2026, including 3.1 Pro and multiple Flash variants. A multi-month slip on a numbered flagship release, confirmed through employee sourcing rather than an official Google announcement, is a departure from that pattern.

Primary sources and further reading

• Google Gemini Launch Delayed as Tech Falls Short of Internal Goals - Bloomberg’s original report by Julia Love and Davey Alba
• Alphabet stock falls 4% on report of Gemini AI model delays - Yahoo Finance on the market reaction
• Gemini 3.5 Pro delays due to coding performance, upgraded Flash model in testing - 9to5Google’s breakdown of the Bloomberg report
• Google Delays Gemini 3.5 Pro Over Coding Issues: Report - Search Engine Journal, including the note that AI Mode still runs on Gemini 3.5 Flash
• Google releases three new Gemini models, but no 3.5 Pro - TechCrunch on the July 21 Flash-only release
• AI researchers continue to leave Google for its rivals - TechCrunch on the June DeepMind departures
• As top talent leaves Google DeepMind, some question if the lab can remain at the forefront of AI development - Fortune on the June 22 stock drop tied to departures
• OpenAI to publicly release GPT-5.6, rolls out conversational AI models - CNBC on the competing GPT-5.6 launch
• Anthropic releases new model, Opus 5 - Axios on Claude Opus 5’s July 24 release