Google announced Gemini 3.7 Flash on 2026-08-13, describing it as its most intelligent workhorse model yet for coding and agents. The release landed roughly three weeks after Gemini 3.6 Flash, which is the detail worth noting on its own: the Flash tier is now shipping on a cadence measured in weeks, not quarters, and the version number is no longer a reliable proxy for how much changed.
The reported gains over 3.6 Flash are large rather than incremental. On FrontierCode 1.1, a production code quality benchmark, Google reports 43.6 percent against 34.6 percent. On DeepSWE v1.1, which measures long-horizon software engineering, it reports 65.3 percent against 49.0 percent. Document comprehension on the GDP.pdf benchmark moves from 22.0 percent to 34.0 percent, and business process work on AutomationBench moves from 17.0 percent to 30.4 percent. All of these are vendor-run figures from the launch post.
Introductory pricing is 0.75 dollars per million input tokens and 3.75 dollars per million output tokens. Google states that regular pricing begins on 2027-01-01 at double those rates, which makes the discount an explicit, dated promotion rather than an open-ended launch price. Availability at announcement spans Google AI Studio, Android Studio and Google Antigravity for developers, the Gemini Enterprise Agent Platform for enterprises, and Gemini Spark for AI Pro and Ultra subscribers across more than 160 countries.
The practical consequence for teams is that the cheap tier is now the tier where agentic coding gets done, and the price is scheduled to double in under five months. Anyone building a cost model on the introductory rate should carry the 2027-01-01 step change in the forecast, and anyone who benchmarked against 3.6 Flash a month ago has a stale result.