Google DeepMind released Gemini 3.7 Flash on 13 August, three weeks after Gemini 3.6 Flash. The cadence is the story as much as the model.
What moved
The largest gains are on agentic and document work. AutomationBench, which scores enterprise workflow automation, rose to 30.4% from 17.0%. DeepSWE v1.1, covering long-horizon software engineering, went to 65.3% from 49.0%. GDP.pdf document comprehension reached 34.0% from 22.0%, FrontierCode 1.1 production code quality 43.6% from 34.4%, and WebDev Arena Elo 1588 from 1538.
The price
Introductory rates are $0.75 per million input tokens and $3.75 per million output, holding until 31 December 2026, then rising to $1.50 and $7.50. Even at the standard rate that is roughly half the previous Flash generation — a price cut and a capability jump in the same release, which is the pattern that has made Flash-class models the default for high-volume agent loops.
Distribution on day one
The model is live in Google AI Studio, Android Studio, the Gemini Enterprise Agent Platform and through Spark for AI Pro and Ultra subscribers in 160+ countries. On the same day GitHub added it to Copilot's model picker across eight surfaces — VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the CLI, the cloud agent and the Copilot app — billed at provider list pricing rather than a premium-request multiplier. Vercel put it on AI Gateway at 50% off.
The caveat on the numbers
All of these benchmarks are vendor-published, run by Google against its own prior model. Independent replication typically lags a release by weeks, and AutomationBench in particular has little external history.