Google Keeps Shipping Flash Models. The One Everyone Waited For Is Still Missing.

Three new Gemini models landed on July 21. None of them is the one Google promised would arrive in June.

Google's release on Tuesday includes Gemini 3.6 Flash, a refined successor to the 3.5 Flash that was the centerpiece of Google I/O in May; Gemini 3.5 Flash-Lite, a cheaper, stripped-down tier for high-volume inference workloads; and Gemini 3.5 Flash Cyber, a specialized model Google built for security teams — its first-ever model dedicated to cybersecurity. The DeepMind blog frames the launch as a response to user feedback on the earlier Flash generation: better code generation, lower latency, tighter token budgets.

What's absent from the announcement is louder than anything in it.

Where Is 3.5 Pro?

Gemini 3.5 Pro was announced at Google I/O 2026 as the company's next frontier model — the large, reasoning-heavy counterpart to the fast-and-cheap Flash line. It was supposed to land in June. June came and went. July is now three weeks old, and Google's public posture has shifted from "soon" to a vague admission that the model is still in testing. Ars Technica reported that Google is now training Gemini 4, which makes the absence of 3.5 Pro feel less like a scheduling delay and more like a strategic recalibration — or a sign of trouble in training the larger model.

The company won't comment on when 3.5 Pro will ship. Meanwhile, Anthropic ships Claude Fable 5 at the frontier. Moonshot's Kimi K3, released last week, prompted another round of "Silicon Valley is shooketh" headlines. OpenAI's latest models continue to define the capability ceiling. Google, by contrast, is shipping efficiency improvements and niche variants.

The Cybersecurity Bet

The most strategically interesting release of the three is Gemini 3.5 Flash Cyber. Built on the 3.5 Flash architecture, it's a model tuned specifically for cybersecurity work — finding vulnerabilities, patching them, and automating security workflows at a fraction of the cost of larger systems. The Verge reports that Google is positioning it as a cheaper alternative to Anthropic's Mythos, which remains one of the most capable — and most expensive — AI security tools on the market.

Flash Cyber is initially available only to governments and trusted partners through CodeMender, Google's security-focused coding agent. The pricing model reflects a deliberate strategy: instead of one massive model that does everything, Google is betting on specialized, cost-efficient models that enterprises can deploy at scale. CodeMender can call Flash Cyber "multiple times at high speed and low cost" per security task, making it economically viable for organizations that couldn't justify the price tag of a Mythos deployment.

A split visual showing a server rack with glowing blue data streams on one side and a green-shielded security operations center on the other, with abstract data flow patterns connecting them

The Flash Line Gets Faster and Cheaper

Gemini 3.6 Flash, the headline release, cuts output token costs by 17% compared to 3.5 Flash and reduces total token usage by up to 65% — a direct response to the rising enterprise anxiety about AI inference costs. The improvements come from architectural refinements rather than a fundamentally new design, which is why Google is calling it a 3.6 rather than a 4.0. For developers building agentic applications — the use case Flash was designed for — the efficiency gains are the difference between shipping a feature and shelving it for budget reasons.

3.5 Flash-Lite, meanwhile, is exactly what it sounds like: a further cost-reduced tier for workloads that don't need the full Flash capability. Think classification, routing, simple extraction — the undramatic but volume-heavy work that makes up most real-world AI usage.

What the Delay Actually Means

Google is telling a coherent story about efficiency, specialization, and cost. It's a good story for enterprise customers who are tired of paying premium token prices for capabilities they don't need. But it's a hard story to square with the competitive reality of the AI industry in mid-2026, where the most attention — and the most developer mindshare — flows to the models at the capability frontier.

The company is teasing Gemini 4, which suggests the long-term roadmap isn't abandoned. But in the short term, the gap between what Google ships and what its competitors ship at the frontier is as wide as it has been since the Gemini brand launched. Efficiency is a valid strategy. It's just not the one that makes headlines — unless the headline is about the model that's still missing.

Google Keeps Shipping Flash Models. The One Everyone Waited For Is Still Missing.

Where Is 3.5 Pro?

The Cybersecurity Bet

The Flash Line Gets Faster and Cheaper

What the Delay Actually Means

Sources

Want to learn more?

Related Articles