Google Launches Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, Delays Pro Model

Google released three new Gemini models on July 21, 2026, expanding its AI lineup with a faster workhorse model, a lightweight variant, and a specialized cybersecurity tool. The company also confirmed that the flagship Gemini 3.5 Pro remains delayed and that it has started pre-training for Gemini 4.

The 3.6 Flash model reduces token usage by up to 65 percent on complex coding tasks compared to its predecessor, according to Josh Woodward, Google's vice president of product. Its cost per task drops roughly 18 percent relative to 3.5 Flash, making it more economical for developers.

Three Models, Three Roles

Each new model targets a different use case. 3.6 Flash is positioned as a general-purpose workhorse with improved coding, knowledge work, and multimodal performance. Meanwhile, 3.5 Flash-Lite is the fastest model to date and 3.5 Flash Cyber is built specifically for vulnerability discovery and patching. Key specifications include:

• 3.6 Flash: Lower price point than 3.5 Flash, better token efficiency, and improved rankings in coding benchmarks such as Frontend Code Arena, where it climbed from 21st to 12th place.
• 3.5 Flash-Lite: Outputs up to 350 tokens per second. Google claims it is more intelligent than the original Gemini 3 model in many cases and matches the cost of the older Gemini 2.5 Flash, which is approaching end of life.
• 3.5 Flash Cyber: Built on 3.5 Flash and integrated with CodeMender, a vulnerability remediation tool. It can identify, validate, and patch security flaws. The model is initially available only to governments and trusted partners through a limited-access pilot.

Delays and Roadmap

Gemini 3.5 Pro, which Google previously demonstrated at I/O 2026, is not yet ready for release. Logan Kilpatrick, a product lead at Google DeepMind, stated on X that the model will ship when it meets quality standards. The delay comes as Alphabet prepares to report earnings on July 22, and investors are expected to question Google's pace relative to rivals OpenAI and Anthropic. Google DeepMind also announced it has started what Kilpatrick called its most ambitious pre-training run yet, for Gemini 4. Early signs suggest the company is pushing for faster, cheaper inference rather than raw intelligence gains. Both 3.6 Flash and 3.5 Flash-Lite are available now in the Gemini app and via API. The cybersecurity model 3.5 Flash Cyber is expected to expand to additional customers later in 2026, though no specific date has been set. Google says 3.5 Pro will follow once testing is complete, but offered no timeline.