Google DeepMind released three new Gemini models on July 21, bolstering its workhorse Flash lineup while leaving a conspicuous gap at the top end of its product stack.
The company shipped Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. The flagship 3.5 Pro model, which Google teased in May as “already being used internally” and expected to roll out “next month,” was notably absent. Bloomberg reported on July 16 that the model had been delayed internally after failing to meet performance targets.
Workhorse upgrades
Gemini 3.6 Flash is positioned as Google’s new workhorse model, promising improved performance on coding, knowledge work, and multimodal tasks while reducing token consumption by up to 17 percent compared to its predecessor, 3.5 Flash. The efficiency gains make it cheaper per query, a critical advantage as enterprise customers deploy AI agents at scale and watch token budgets balloon.
“The focus on these releases is to deliver efficiency, latency, and reliability to customers that are building AI agents at scale,” Google DeepMind said in the announcement.
Gemini 3.5 Flash-Lite sits at the bottom of the tier, offering the lowest-cost option for production deployments that prioritize budget over capability. Both models are available immediately on Google AI Studio and Vertex AI.
A cybersecurity specialist
The most notable addition is Gemini 3.5 Flash Cyber, a specialized model fine-tuned for finding and fixing cybersecurity vulnerabilities. Unlike the other two models, which are broadly available, Flash Cyber will be offered exclusively to governments and trusted partners through a limited-access pilot program.
The model enters a competitive space that has seen rapid activity in recent months. OpenAI launched GPT-5.5-Cyber in June as part of its Patch the Planet initiative, focusing on open-source vulnerability remediation. Anthropic’s Fable 5, before its export-control entanglement with the US government, demonstrated significant autonomous hacking capabilities. Google’s entry aims to give allied governments and critical infrastructure operators a tool purpose-built for defensive security work.
One model missing
The continued absence of Gemini 3.5 Pro is becoming difficult to ignore. Since Google’s last flagship Pro update in February, OpenAI has shipped GPT-5.5 and begun rolling out GPT-5.6, while Anthropic has released Claude Opus 4.8 and Claude Sonnet 5, and expanded access to its frontier Fable 5 model.
Logan Kilpatrick, product lead at Google DeepMind, confirmed on X that the company is testing Gemini 3.5 Pro with partners and hopes to “land soon.” He also disclosed that the team has started “the most ambitious pre-training run yet” for Gemini 4, suggesting Google is investing heavily in its next-generation architecture rather than iterating on the current one.
The strategy mirrors a pattern the industry has seen before: ship cheaper, faster models to capture production workloads while working toward a step-change improvement in the next generation. Whether that bet pays off depends on how long “soon” turns out to be.
Sources: Google releases three new Gemini models , but no 3.5 Pro (TechCrunch, July 2026); Google reveals faster and cheaper Gemini 3.6 Flash, says 3.5 Pro is still in testing (Ars Technica, July 2026)