Google Unveils Three New Gemini AI Models to Super‑Charge Its Cloud Business

At a breezy press‑event streamed from its Mountain View campus on July 21, Google rolled out three new flavors of its Gemini family. The move wasn’t just a product splash; it’s a clear signal that the company wants to turn its AI‑powered models into a cash‑generating engine for Google Cloud.

First up is Gemini 3.6 Flash, billed as the new “workhorse.” According to the DeepMind team, it shaves roughly 17 % off the token count you’d need compared with the previous 3.5 Flash, meaning you get more output for the same prompt. Pricing follows suit – $1.50 per million input tokens and $7.50 per million output tokens, a noticeable dip from the $9 / M output price of its predecessor.

For customers who run massive workloads on a shoestring budget, Google introduced Gemini 3.5 Flash‑Lite. This version trades a little raw speed for a drastic price cut: $0.30 per million input tokens and $2.50 per million output tokens, while still churning out about 350 tokens per second. It’s the kind of model you’d slot into batch‑processing pipelines or daily report generators without blinking an eye at the bill.

Then there’s the niche offering, Gemini 3.5 Flash‑Cyber. Built specifically for vulnerability‑detection and remediation, it’s being piloted with a handful of government agencies and trusted partners. While it isn’t publicly available yet, Google says it can spot and help fix security holes at a level that rivals Anthropic’s Claude Mythos.

What’s missing, however, is the much‑rumoured Gemini 3.5 Pro. Both TechCrunch and Ars Technica noted that the Pro iteration is still in internal testing, with a launch pushed back to an undefined future date. In the meantime, Google’s product lead for Gemini, Logan Kilpatrick, hinted that the company is already looking ahead to Gemini 4, which is expected to begin pre‑training later this year.

Beyond the specs, the timing feels purposeful. Google is in the middle of a “key cloud growth test,” a phrase the company uses to describe its effort to convert AI buzz into real‑world cloud revenue. By offering cheaper, more token‑efficient models, Google hopes to entice enterprises – especially those already on Google Cloud – to shift AI workloads from competitors to its own infrastructure.

Early benchmark numbers are encouraging. Ars Technica reported a 49 % improvement on the DeepSWE security benchmark (up from 37 % with Gemini 3.5) and an 83 % score on the OSWorld exam, nudging past the 78.4 % mark set by the prior generation. Those gains, combined with the lower price tag, could be enough to tip the scales for data‑heavy firms weighing up OpenAI, Anthropic, or Microsoft alternatives.

All told, Google’s Gemini refresh feels less like a routine upgrade and more like a strategic push to make AI a profit center for its cloud division. Whether the market bites will depend on how quickly enterprises can trust the new models – and whether the still‑mysterious Gemini 3.5 Pro ever sees the light of day.