Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens, down from 3.5 Flash's $9.00 output rate, while using 17% fewer output tokens on the Artificial Analysis Index and up to 65% fewer on the DeepSWE coding benchmark. The efficiency gains matter more than the sticker price suggests: fewer tokens per completed task compounds with the lower per-token rate, so the effective cost drop on agentic coding workloads is larger than 17%. Benchmark scores improved alongside the price cut — DeepSWE rose from 37% to 49%, MLE Bench from 49.7% to 63.9%, and OSWorld-Verified from 78.4% to 83.0%.
The lineup's third model, Gemini 3.5 Flash Cyber, is a specialized system for finding and fixing software vulnerabilities, offered only on a limited-access basis to governments and trusted partners rather than through public API. Google frames it as a defensive tool for security research, distinct from the general-purpose Flash models. None of the three new releases is a flagship: Gemini 3.5 Pro, which Google told the market in May would ship "the following month," remains in partner testing only, with no public timeline. That gap has become the more closely watched story — a single Bloomberg report about the Pro delay erased roughly $200 billion from Alphabet's market value on July 16.
Google confirmed alongside the Flash launch that it has started "our most ambitious pre-training run yet, for Gemini 4," the clearest signal yet that the company's roadmap is bifurcating: ship efficient, cheaper Flash-tier models on schedule for production agentic workloads, while the frontier-reasoning flagship takes longer than promised.
