Google released Gemini 3.7 Flash on August 13, pitching it as the company’s most capable “workhorse” model yet for coding and agent workflows. It landed just three weeks after Gemini 3.6 Flash, which is a fast turnaround even by this year’s standards, and it’s already rolling out across the Gemini API, Google AI Studio, Android Studio, and the Gemini Enterprise Agent Platform.
What actually changed in Gemini 3.7 Flash
According to Google’s own announcement, the gains are concentrated in software engineering and agentic tasks rather than general chat quality. On the DeepSWE v1.1 benchmark, scores moved from 49.0% to 65.3%. On FrontierCode 1.1 Main, a benchmark for shipping working code on the first try, the jump was from 34.4% to 43.6%. Document comprehension and business-automation benchmarks improved as well, though by smaller margins.
In plain terms: fewer half-broken pull requests from the model, better multi-step planning when it’s acting as an agent, and more reliable tool calling when it’s wired into something like Google Workspace.
Pricing and where to get it
Google is running introductory pricing through the end of 2026 at $0.75 per million input tokens and $3.75 per million output tokens, roughly half of what 3.6 Flash cost. Standard pricing of $1.50 and $7.50 per million tokens kicks in on January 1, 2027. It’s available now through Google Antigravity, the Gemini API, Android Studio, and for Google AI Pro and Ultra subscribers via Spark in the Gemini app.
Who should care
If a team is already building on Gemini for a coding agent, an internal tool, or a client-facing AI feature, this is a straightforward model swap worth testing against the existing prompts before committing, since benchmark gains don’t always translate cleanly to a specific use case. The pricing cut also matters for anything token-heavy, like a support bot that processes long documents or an agent that reads through a codebase before making changes. We run into this kind of evaluation regularly on AI Development engagements, where the right model for a client often changes every few months whether anyone asked for it to or not.
