Google Releases Gemini 3.6 Flash and 3.5 Flash-Lite for Agent Workloads
Google is positioning two new Gemini Flash models around lower-cost, lower-latency agent workloads, while keeping its cyber-focused variant in a limited-access pilot.
Google announced Gemini 3.6 Flash and Gemini 3.5 Flash-Lite on July 21, making both models available through its developer and enterprise channels. The release is aimed at teams building agentic systems that need a balance of capability, latency and operating cost. Google is also introducing a specialized Gemini 3.5 Flash Cyber model, but says that variant will initially be limited to governments and trusted partners through its CodeMender security-agent pilot.
What changed
Gemini 3.6 Flash is Google's new general-purpose Flash model for coding, knowledge work and multimodal tasks. Google says it uses fewer output tokens than Gemini 3.5 Flash and lists lower token prices for the newer model. Those are vendor claims, not an independent comparison, but they clarify the commercial position: Google is competing on the cost of completing an agent task, rather than solely on a headline benchmark.
Gemini 3.5 Flash-Lite is the faster, high-throughput sibling. Google describes it as intended for workloads such as agentic search and document processing, where low latency and large request volumes matter. It is available through the Gemini API and Google AI Studio, as well as the company's enterprise products.
The cyber-focused model has a materially narrower release path. Google says Gemini 3.5 Flash Cyber is tuned to find and fix software vulnerabilities and will be offered through a limited-access CodeMender pilot. The company frames the restriction as a safeguard for technology it considers dual-use.
Why it matters
The launch separates everyday agent infrastructure from more sensitive security automation. Developers can use the two general Flash models now, while the cyber model's availability is tied to a controlled pilot. That distinction is important for organizations evaluating whether a model announcement represents immediate access or a future capability with eligibility constraints.
Google's performance and efficiency descriptions are its own measurements and should be read as product claims until independently tested. The more durable signal is the product split: a general model for scaled agent workflows, a lower-cost variant for throughput, and a restricted model for vulnerability work.
Status
Confirmed. Google published the announcement and availability details through its official newsroom. Internal confidence is high because Google is the primary authority for this product release; the performance claims remain vendor-reported.
Sources
Update note: Last reviewed 2026-07-29. We will revise this post if Google changes availability, pricing, or the scope of the limited-access cyber pilot.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.