Google Skips Ahead to Gemini 4 Training While 3.5 Pro Stays Delayed

Google released three new Gemini models on Tuesday: Gemini 3.6 Flash, 3.5 Flash-Lite, and a government-only cybersecurity variant called 3.5 Flash Cyber. Gemini 3.5 Pro, the flagship model Google promised for June, was not among them.

Logan Kilpatrick, Google DeepMind’s product lead, said Pro is being tested with partners and should land soon. He added, almost in passing, that the team has already begun its most ambitious pretraining run yet, for Gemini 4, a model whose existence Google has barely discussed until now.

Three Flash Models Ship, With One Reserved for Governments

Gemini 3.6 Flash is what Google calls its workhorse model, built for coding, knowledge work and multimodal tasks. According to Google’s own developer blog, it cuts output token usage by 17% compared with 3.5 Flash, while taking fewer reasoning steps and tool calls to finish multi-step jobs. That efficiency gain also makes it cheaper to run than the model it replaces.

Gemini 3.5 Flash-Lite is the cheapest and fastest model in the lineup, aimed at high-volume, latency-sensitive workloads. Gemini 3.5 Flash Cyber is different in kind: a version of 3.5 Flash fine-tuned specifically to find and patch software vulnerabilities. In stress tests on complex codebases, DeepMind said the model went surpassing Claude Opus 4.6 on Chrome and Safari vulnerability tests, outperforming both its own mainline Flash models and Anthropic’s coding-focused model.

Google’s push into AI-driven vulnerability hunting arrives while its own Android assistant has had security headaches of its own, including repeated lock screen bypasses tied to Gemini features. That contrast is not lost on security researchers watching Google position Flash Cyber as a defender’s tool.

A Flagship That Keeps Missing Its Own Deadlines

Back in May, Google said Gemini 3.5 Pro was “already being used internally, and we look forward to rolling it out next month.” June came and went. On July 16, Bloomberg reported that Pro was months behind schedule, having fallen short of internal performance goals, especially in coding.

Alphabet shares fell 4% the day that report published, CNBC reported, as investors weighed the cost of a stalled flagship against a field moving fast in every direction. Bloomberg’s sourcing pointed to a late-June refresh of the model’s training data that underperformed expectations, plus friction among the multiple internal teams that sign off on a model before release.

We’re currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, and we’re productively engaged with the U.S. government on model testing and broader frameworks.

A Google spokesperson gave that statement to Bloomberg in response to the delay report. Search Engine Journal’s earlier coverage, also citing Bloomberg, noted senior departures from Google’s AI organization amid internal concern about the company’s coding tools falling behind rivals.

What we know:

• Google shipped Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber on Tuesday, July 21.
• Bloomberg reported the Pro delay traces mainly to coding performance shortfalls.
• Kilpatrick says Pro is in partner testing now and Gemini 4 pretraining has started.
• Alphabet shares dropped 4% on the day the delay report broke.

What’s unconfirmed:

• No public ship date exists for Gemini 3.5 Pro.
• Whether Gemini 4 arrives close enough to 3.5 Pro to make Pro a short-lived release.
• Any specifications for Gemini 4 beyond the fact that pretraining is underway.

The Cadence Google Broke

Until this year, Google refreshed its Pro tier roughly every three months. That rhythm is why a July delay reads differently than an ordinary product slip.

1. November 18, 2025: Gemini 3 Pro launches as Google’s most intelligent model at the time.
2. February 19, 2026: Gemini 3.1 Pro succeeds it, about three months later, holding pricing steady while improving reasoning and long context performance.
3. May 19, 2026: Google unveils Gemini 3.5 Flash at I/O and says Pro is already used internally, expected the following month.
4. June 2026: The month Google named for Pro’s rollout passes without a release.
5. July 16, 2026: Bloomberg reports Pro is months behind schedule on coding performance.
6. July 21, 2026: Google ships three Flash-tier models instead, and Kilpatrick says Pro testing continues.

By that count, Google has gone roughly five months since its last Pro update, well past the three-month gap between 3 Pro and 3.1 Pro. The company has never let that gap run this long before.

Why Start Gemini 4 Before Finishing Gemini 3.5?

Google has not explained the overlap directly. But Kilpatrick’s Tuesday comments put a live pretraining run for Gemini 4 on the record before Gemini 3.5 Pro has reached a public release, which points to a company more focused on leapfrogging its current generation than perfecting it.

Labs typically stay quiet about a next model until the current one ships. Kilpatrick did the opposite, saying the team is testing 3.5 Pro with partners now while separately confirming its most ambitious pretraining run yet is already underway for Gemini 4. That means Gemini 4’s development is running alongside Gemini 3.5 Pro’s delayed cleanup, not after it.

If the coding shortfalls behind the Pro delay are structural rather than a scheduling hiccup, the Pro model that eventually ships risks landing as a bridge release, arriving just as attention and internal resources shift toward Gemini 4. Google has not said that outright. Its own timeline says it plainly enough.

Coding Agents Are Where Rivals Already Have the Edge

CNBC reported that Pro’s coding shortfall comes as rivals including OpenAI have recently debuted models that outpace Google’s current offerings in generating software code. In the time since Gemini Pro was last updated, OpenAI has released GPT-5.5 and begun rolling out GPT-5.6, while Anthropic has launched Claude Opus 4.8 and Claude Sonnet 5 and expanded access to its frontier Fable 5 model.

DeepMind staff have internally flagged a lack of a credible commercial coding product for businesses building AI development tools, the exact segment where OpenAI’s Codex and Anthropic’s Claude Code have built early leads, according to Bloomberg’s reporting. Regulators are adding pressure from a different angle: the European Union has already ordered Android’s assistant slot open to rivals like ChatGPT and Claude, chipping at the distribution advantage that has cushioned Gemini so far.

Google’s model quality now carries stakes outside its own apps, too. Gemini also powers the new Siri experience Apple shipped on Apple Watch this year, meaning a weak link in Google’s model lineup is no longer just Google’s problem to manage quietly.

What Happens if Gemini 4 Slips Too

Two outcomes sit ahead of Google, and only one of them is good. If Gemini 4’s pretraining run avoids the coding regressions that stalled Pro, Google could leapfrog straight into a stronger generation and make the current delay a footnote. If the same performance ceiling reappears, Google will have spent months chasing a moving target twice in a row while OpenAI and Anthropic keep shipping.

Enterprise customers building AI agents are not waiting to find out. They are building on Flash today, cost and latency being the selling points Google can actually deliver right now.

For now, the only Gemini tier running at full strength is Flash. Pro remains a promise. Gemini 4 remains a training run.

Frequently Asked Questions

What is Gemini 3.6 Flash, and how is it different from 3.5 Flash?

Gemini 3.6 Flash is Google’s workhorse model for coding, knowledge work and multimodal tasks, priced at $1.50 per million input tokens and $7.50 per million output tokens, down from $9 per million output tokens for 3.5 Flash. It is also available inside Google Antigravity and the Gemini Enterprise app, in addition to AI Studio and Android Studio.

Is Gemini 3.5 Pro available to the public yet?

No. As of press time, Google’s developer platforms still list gemini-3.1-pro-preview rather than a public gemini-3.5-pro model ID. A July 17 target that circulated in trade reports came and went without an official release, and Google has given no new date beyond Kilpatrick’s comment that the model should land soon.

Who can use Gemini 3.5 Flash Cyber?

Access is limited to governments and trusted partners through the CodeMender pilot program. Google says the model was stress-tested on Google Chrome’s production commit-scanning pipeline using vulnerabilities that were never publicly disclosed, keeping the benchmark free of contamination from prior training data.

What is Gemini 4, and when will it launch?

Google has confirmed only that it has started what Kilpatrick called its most ambitious pretraining run yet for Gemini 4. No timeline has been given, and pretraining is typically the earliest phase of model development, arriving many months before a model reaches even internal testing.

Why did Google move from 3.5 Flash to 3.6 Flash so quickly?

Gemini 3.6 Flash arrived roughly two months after 3.5 Flash’s May debut, a far faster refresh than Google has ever used on its Pro tier. It shows Google iterating quickly on its cheaper production model even as the flagship stalls behind it.