Google Rolls Out New Gemini Models, Targeting Cheaper, Faster AI for Cloud and Security

On July 21, 2026, a day before Alphabet’s earnings call, Google announced a modest but strategic expansion of its Gemini family. Three new flavors—Gemini 3.5 Flash‑Cyber, Gemini 3.6 Flash, and Gemini 3.5 Flash‑Lite—hit the market together, each promising a different slice of the AI‑as‑a‑service pie.

First up is Gemini 3.5 Flash‑Cyber, a model built expressly for hunting software bugs and patching vulnerabilities. Google is being cautious, limiting early access to governments and a handful of “trusted partners.” The company says the new security‑focused engine costs less per token than its larger siblings, a claim that, for now, rests on internal benchmarks.

Next comes Gemini 3.6 Flash, which Google touts as a more efficient all‑rounder. According to the firm, it can accomplish the same tasks while using up to 17 % fewer tokens, translating into lower per‑token pricing. The model reportedly shines on coding assistance, multimodal queries, and the kind of knowledge‑work that powers AI‑driven agents.

Rounding out the trio is Gemini 3.5 Flash‑Lite, billed as the fastest and cheapest option in the 3.5 series. It’s aimed at high‑volume, lightweight workloads—think the repetitive prompts that flood AI‑agent pipelines every day.

Google isn’t just adding variants for the sake of novelty; it’s a clear response to mounting competition. Anthropic’s new cybersecurity model, Mythos, is the most direct rival to Flash‑Cyber. Meanwhile, Chinese challengers such as Moonshot AI’s Kimi K3 and Alibaba’s Qwen 3.8 Max—both said to lag behind Anthropic’s Fable 5—are sharpening the global AI arms race.

In its press materials, Google even goes as far as to claim the new Gemini models are “cheaper per task” than OpenAI’s GPT‑5.6 Terra Max, Kimi K3, and Qwen 3.7 Max. While those numbers sound impressive, they’re based on Google’s own calculations and have yet to be validated by independent testing.

All told, the rollout is a calculated gamble: offer lower‑cost, higher‑efficiency models to keep cloud customers happy and to signal that Google’s AI stack can hold its own against a crowded field of rivals. Whether the market buys into the token‑saving narrative remains to be seen, but the timing—right on the heels of Alphabet’s earnings—suggests the company sees these models as a key piece of its next growth chapter.