Google has expanded its Gemini lineup with three new models, unveiled on July 21, 2026, each tuned for a very different kind of job. The headliner is Gemini 3.6 Flash, which Google positions as its all-round workhorse — the model most people and businesses will actually run day to day.
The pitch for 3.6 Flash is efficiency without a quality trade-off. Google says it delivers stronger coding and knowledge-work performance while chewing through fewer resources: on the Artificial Analysis Index, it consumes 17% fewer output tokens than the previous 3.5 Flash. Fewer tokens means lower bills, which matters at scale. Speaking of bills, pricing lands at $1.50 per million input tokens and $7.50 per million output tokens.
Where 3.6 Flash really flexes is multimodal work. Google says customers have found it particularly capable at parsing documents, analysing charts and data, and drafting reports — the unglamorous but genuinely useful tasks that eat up office hours.
Sitting below it is Gemini 3.5 Flash-Lite, the speed demon of the group. It’s built for low-latency and high-throughput jobs — think agentic search and bulk document processing where volume, not deliberation, is the priority. It clocks in at 350 tokens per second and is priced to move at $0.30 per million input tokens and $2.50 per million output tokens. That combination makes it an obvious pick for developers running large pipelines who need answers fast and cheap.
Both 3.6 Flash and 3.5 Flash-Lite are already available across the board: developers can reach them through the Gemini API in Google AI Studio, businesses via Gemini Enterprise, and everyone else through the Gemini app.
The third model is the intriguing one. Gemini 3.5 Flash Cyber is a specialist, trained specifically to find, verify and fix security vulnerabilities in code. Rather than being a general assistant that happens to know about security, it’s aimed squarely at the defensive side of software — spotting flaws and proposing fixes before attackers get there.
Unlike its siblings, Flash Cyber isn’t open to the public. For now it’s restricted to government agencies and select partners, and Google hasn’t published pricing for it. That gated rollout makes sense: a model purpose-built to probe code for weaknesses is exactly the kind of tool you’d want to keep on a short leash.
Taken together, the three releases sketch out Google’s current strategy neatly. There’s a capable generalist for the mainstream, a lean and rapid option for developers who prioritise throughput, and a locked-down specialist for one of the highest-stakes corners of the industry. It’s less a single flagship launch than a deliberate carving-up of the workload, with each model earning its keep in a specific lane.
