Google's Gemini 3.7 Flash launched at an introductory price roughly half its predecessor's original cost while posting sharp gains on coding and business-automation benchmarks, changing which workflows are cheap enough to hand to an agent.
Gemini Spark, Google's always-on personal agent, spends its day consolidating files, drafting emails, and updating status documents so a person doesn't have to. Every one of those actions bills against a token meter, and until this week that meter ran twice as fast. On Wednesday, Google introduced Gemini 3.7 Flash, the model Spark switched to the same day, at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, which Google describes as half of what its predecessor originally cost.
3.7 Flash shipped just three weeks after Gemini 3.6 Flash, and Google frames the release as a direct response to developer feedback rather than a from-scratch model cycle. The gains land where a business would actually feel them. On FrontierCode 1.1, a coding benchmark, the score rose from 34.4 percent to 43.6 percent. On DeepSWE, which measures debugging and issue resolution, it climbed from 49.0 percent to 65.3 percent. On AutomationBench, which Google says tests a model's ability to complete real-world business workflows, the jump was from 17.0 percent to 30.4 percent. Google also reports a gain on the GDP.pdf benchmark, an eval for processing complex documents in finance, law, and biosciences, from 22.0 percent to 34.0 percent. In web development, Google says 3.7 Flash generates more functional layouts and feature-complete apps in fewer prompts, and it outperforms 3.6 Flash on Arena.ai's WebDev Arena with an Elo score of 1588 versus 1538.
The pricing checks out against Google's own Gemini API pricing page: $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, reverting to $1.50 and $7.50 on January 1, 2027. One wrinkle worth flagging: that same page currently lists Gemini 3.6 Flash at the identical promotional rate, so the "half the cost" claim in Google's announcement is a comparison against 3.6 Flash's original launch price, not against what either model bills today.
For a marketing or ops leader, the model name is not the point. The point is what happens to the math on which workflows are worth automating. A coordinator who routes requests, fills in forms, and chases status updates all day is doing roughly what AutomationBench is built to measure, and a model that gets almost twice as good at that job while costing half as much per token changes which of those jobs clear the bar for handing to an agent instead of a person. The same logic applies to anyone running a coding agent against a ticket queue: FrontierCode and DeepSWE are both proxies for the kind of unattended bug fixing and pull request generation that used to require a developer's attention on every pass. And for teams standing up internal tools without pulling engineering off the roadmap, the WebDev Arena gain matters in a very literal sense: fewer prompts to get a usable app means less time paying for a model's mistakes before the thing actually works.
Access is broad rather than gated. Developers can reach 3.7 Flash through Google Antigravity, the Gemini API in Google AI Studio, or Android Studio. Enterprises get it through the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Consumers already have it by default, since Gemini Spark switched over for Google AI Pro and Ultra subscribers in more than 160 countries starting the day of the announcement.
All of the benchmark numbers above are Google's own, measured on Google's own evals, with no independent replication published yet. The discount is also a countdown, not a new baseline: it expires at the end of the year, and the rate then doubles. Anyone budgeting an agent workflow around 3.7 Flash's current price should plan for that reversion rather than assume it holds.
A model that gets meaningfully better at business automation while its price gets cut in half is not a headline that announces itself as a labor story, but that is what it is. Nobody hires a coordinator by press release. They just quietly stop opening the requisition.