Google launches Gemini 3.7 Flash, its fastest coding and agent model yet
Google has released Gemini 3.7 Flash, the latest entry in its low-latency Flash line, positioning it as the company's strongest model yet for software development and autonomous agent work.
What's new
In its announcement, Google says "3.7 Flash is our most intelligent workhorse model yet for coding and agents." The company backs that claim with benchmark gains over its predecessor:
- FrontierCode 1.1 Main: 43.6% accuracy, up from 34.4% for Gemini 3.6 Flash
- DeepSWE v1.1: 65.3%, up from 49.0% for the previous version
- Improvements in complex document processing and business-workflow automation
On pricing, Google is running an introductory rate through December 31, 2026: $0.75 per million input tokens and $3.75 per million output tokens. Standard pricing takes effect January 1, 2027, at $1.50 per million input tokens and $7.50 per million output tokens — double the introductory rate.
The model is available now through Google AI Studio, Android Studio, Gemini Enterprise, and Gemini Spark for Pro and Ultra subscribers.
Context
Flash has been Google's answer to the industry's push toward cheaper, faster models that can still handle real coding and agentic workloads, sitting below the flagship Gemini Pro/Ultra tier on cost while narrowing the capability gap. The 3.6-to-3.7 jump continues a pattern of frequent Flash refreshes rather than a single annual release, mirroring how OpenAI and Anthropic have been iterating on their own lightweight tiers. The FrontierCode and DeepSWE gains in particular target the agentic coding use case — multi-step tasks where a model has to plan, edit, and verify code changes rather than just complete a snippet.
Why it matters
The two-tier pricing structure — a steep introductory discount that roughly doubles in five months — signals Google's confidence that developers will lock in usage now and tolerate the increase later, a strategy also used elsewhere in the industry (Anthropic recently reversed a similar planned increase for Sonnet 5). For teams building coding agents, a workhorse-tier model that meaningfully narrows the gap to frontier performance on FrontierCode and DeepSWE benchmarks while staying cheap is a direct cost lever: more of an agentic pipeline can run on the fast, inexpensive tier rather than escalating to the flagship model. The introductory pricing window also gives Google roughly four months to capture switching behavior before the price effectively doubles.