Google releases a cheaper new Gemini model aimed at coding agents

Google DeepMind released Gemini 3.7 Flash, an update to its mid-tier Flash line pitched at coding and agentic workflows, three weeks after it shipped Gemini 3.6 Flash. Google priced the new model at $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, after which the rate rises to $1.50 and $7.50 — roughly half the introductory price of 3.6 Flash, and a discount Google framed as intended to encourage developers to build agentic coding products on the tier rather than reach for a more expensive flagship model.

Google reported benchmark gains alongside the price cut: accuracy on the FrontierCode production-code benchmark rose from 34.4% to 43.6%, and on DeepSWE from 49.0% to 65.3%, figures the company said reflected better instruction-following and fewer retries in agentic coding sessions. These are Google’s own reported numbers rather than independently verified scores.

The release continued a pattern of rapid iteration on Google’s lower-cost Gemini tier through mid-2026, arriving weeks after 3.6 Flash and its siblings while Google’s flagship Gemini 3.5 Pro remained unreleased, still described as being tested with partners. Frequent, aggressively priced updates to the efficiency tier — rather than a flagship release — have been Google’s primary lever against rival labs’ coding-agent products through this period.