AI Models

Google Ships Three New Gemini Models — And Still No 3.5 Pro

Newaiera Desk · 2026-07-22 · 4 min read

Google DeepMind released Gemini 3.6 Flash, 3.5 Flash-Lite and a cybersecurity-tuned Flash Cyber on July 21. The flagship Pro model that Sundar Pichai promi

Google DeepMind shipped three models on Tuesday. None of them was the one everyone was waiting for.

On July 21, the company released Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber — a trio aimed squarely at the unglamorous economics of running AI agents in production. What it did not release, again, was an update to Gemini Pro, its flagship reasoning tier, which has not been refreshed since February.

What actually shipped

Gemini 3.6 Flash is positioned as the everyday workhorse. Google says it improves on coding, knowledge work and multimodal performance while cutting token usage by up to 17% compared with its predecessor, Gemini 3.5 Flash. Fewer tokens for the same job means a lower bill, which is the entire pitch.

Gemini 3.5 Flash-Lite goes further down the same road: Google describes it as the most cost-effective model in its class.

Gemini 3.5 Flash Cyber is the odd one out and the most interesting. It is a specialised variant fine-tuned to find and fix security vulnerabilities. Notably, it is not going on general release — Google is restricting it to governments and trusted partners through a limited-access pilot programme. That is a meaningful signal in itself: a model good enough at finding exploitable bugs that its maker would rather not hand it to everyone.

The common thread across all three is efficiency, latency and reliability for customers building agents at scale. This is not a frontier-capability release. It is an infrastructure release.

The Pro-shaped hole

The absence is the story.

Gemini Pro is Google's high-capability tier — the one built for complex reasoning and hard coding problems, where Flash models trade capability for speed and cost. It was last updated in February.

Back in May, alongside the Gemini 3.5 Flash launch, Google said the Pro version was already in internal use and would roll out the following month. June came and went. On July 16, Bloomberg reported that the launch had slipped because the model was falling short of Google's own internal performance targets.

Google DeepMind product lead Logan Kilpatrick said on Tuesday that 3.5 Pro is currently being tested with partners and that the team hopes it will land soon. In the same breath, he noted something arguably bigger: DeepMind has begun its most ambitious pre-training run yet — for Gemini 4.

That is a genuinely awkward pair of statements. The flagship you promised two months ago is still in partner testing, but the generation after it is already cooking.

Why five months matters in 2026

In a normal software market, a one-quarter delay on a flagship is a rounding error. This is not a normal software market.

Since Gemini Pro's February update, OpenAI has released GPT-5.5 and started rolling out GPT-5.6. Anthropic has launched Claude Opus 4.8 and Claude Sonnet 5, and expanded access to its frontier Fable 5 model. The competitive bar Google has to clear on arrival keeps rising while it waits.

There is a defensible strategy hiding in here. Most production AI spend is not on frontier reasoning — it is on high-volume, latency-sensitive, cost-sensitive agent workloads, which is exactly what Flash, Flash-Lite and Cyber are built for. Shipping cheaper tokens to the workloads that actually generate revenue is not a consolation prize.

But flagship models set the perception of who is ahead, and perception drives enterprise procurement. Every month the Pro tier stays dated is a month those evaluations get resolved somewhere else.

What to do with this

If you are running agents at volume, 3.6 Flash is worth benchmarking against your current routing this week — a 17% token reduction compounds fast at scale.

If you are waiting on 3.5 Pro before committing to a frontier vendor, note that Google still has not published a model card, pricing or a firm date. Treat any specific launch date or benchmark figure you see circulating as speculation until Google publishes one.

And keep an eye on the Gemini 4 pre-training comment. It is the most concrete thing Google has said in weeks about where its real attention is.