Google DeepMind releases three new Gemini models focused on efficiency and cybersecurity

Google DeepMind released three new Gemini models on July 21, 2026: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. The models are positioned as cost-efficient, high-performance options for building AI agents at scale, with Google emphasizing improvements in coding, knowledge work, and multimodal performance.

Gemini 3.6 Flash is described as Google’s "workhorse model," reducing token usage by up to 17% compared to its predecessor, 3.5 Flash, while maintaining competitive capabilities. Google claims this makes it cheaper to operate for production applications.

Gemini 3.5 Flash-Lite is positioned as the most cost-effective model in its class, targeting scenarios where price sensitivity is critical. In contrast, 3.5 Flash Cyber is a specialized model fine-tuned for identifying and addressing cybersecurity vulnerabilities, available exclusively to governments and trusted partners through a limited-access pilot program.

The releases omit the long-anticipated update to Google’s flagship model, Gemini 3.5 Pro, which was last updated in February 2026. Google had previously teased its release in May 2026, stating it was already in internal use and expected to launch the following month. However, Bloomberg reported last week that internal delays have slowed progress due to performance challenges.

Google DeepMind product lead Logan Kilpatrick stated that the company is currently testing Gemini 3.5 Pro with partners and expects it to "land soon." He also noted that the team has initiated its most ambitious pre-training run yet for the next-generation Gemini 4.

The new models reflect a strategic shift toward efficiency, latency, and reliability for production-scale AI agents, while the absence of 3.5 Pro highlights competitive pressures from rivals like OpenAI and Anthropic, which have recently released or expanded access to advanced models such as GPT-5.5, GPT-5.6, Claude Opus 4.8, and Claude Sonnet 5.