Google’s Gemini 3.7 Flash cuts pricing in half with big coding gains

Shipping a major model update usually takes months. Google just did it in twenty-one days.

Google officially rolled out Gemini 3.7 Flash, its newest workhorse model designed specifically for software engineering and agentic workflows. Even better: while performance across core tasks jumped significantly, input pricing was cut squarely in half to $0.75 per million tokens.

Sharper coding, faster execution

With reports noting that Google delayed its promised Gemini 3.5 Pro, the team clearly redirected its firepower into upgrading its high-volume workhorse.

On the developer front, 3.7 Flash delivers massive jumps on real-world coding benchmarks. Its score on DeepSWE v1.1 climbed from 49.0% to 65.3%, while FrontierCode 1.1 Main rose from 34.4% to 43.6%. In plain English: it writes cleaner production code on the first pass and gets stuck far less often during complex bug fixes.

Document reasoning and workflow automation saw similar leaps. The model scored 34.0% versus 22.0% on the GDP.pdf evaluation and jumped from 17.0% to 30.4% on AutomationBench. For front-end builders, it scored 1588 on WebDev Arena.

Early hands-on reviews also point out its blazing throughput. Independent testing clocked its generation speed at around 340 output tokens per second, making multimodal tasks like generating interactive web layouts from PDFs feel practically instantaneous.

Half the price, but still behind a paywall

Raw horsepower is nice, but pricing is where 3.7 Flash really turns heads.

Through the end of the year, Google is offering an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens. That cuts the original token costs of 3.6 Flash right in half, explicitly targeting developers looking to run multi-step agents at scale without breaking the bank.

Access for everyday individuals, however, remains limited. Casual users will not see it widely unlocked in free tiers just yet.

Developers can jump in today via Google AI Studio, Android Studio, and Google Antigravity, while enterprises get access through Gemini Enterprise. For individual power users, it is exclusively available inside the Gemini app for Google AI Pro and Ultra subscribers across 160 countries, powering their always-on personal agent, Gemini Spark.

Why it matters

If you build software or rely on multi-step agents, this release is an immediate win. Running autonomous loops requires relentless tool calls and self-correction, which normally burns through token budgets. Getting a 65.3% DeepSWE score at $0.75 per million input tokens with 340 tokens per second makes building production agents substantially cheaper and less prone to manual intervention.

For the AI industry, Google is clearly shifting its competitive focus. While frontier flagship models remain wildly expensive to run at scale, Google is weaponizing its lightweight Flash tier. Dropping a major coding upgrade at half price just three weeks after its predecessor puts immense pricing pressure on rival small models.

We think this is a pragmatic move by Google. Even as delays to Gemini 3.5 Pro suggest potential friction at the very top tier, doubling down on the workhorse layer hits where developers actually spend money. Prioritizing disciplined execution, document reasoning, and aggressive pricing makes 3.7 Flash one of the most practical production models available right now.