Glonce

Google ships Gemini 3.7 Flash at half the price of 3.6 Flash

modelsThe Glonce editorsHacker News🔥80Glonce rating80Impact72Novelty48Relevance100Surprise42Players92Credibility87
Google ships Gemini 3.7 Flash at half the price of 3.6 Flash

Google has released Gemini 3.7 Flash, which it calls its most intelligent workhorse model yet for coding and agents. The release lands just three weeks after Gemini 3.6 Flash, and Google says it is a direct result of developer feedback and algorithmic innovations the company looks forward to bringing to future models.

On coding, Google reports 3.7 Flash improving on debugging, issue resolution, and first-pass code accuracy, plus better production-ready code generation: it scores 43.6% versus 3.6 Flash's 34.4% on the FrontierCode 1.1 Main benchmark, and 65.3% versus 49.0% on DeepSWE v1.1. In web development, Google says the model produces more functional layouts and feature-complete apps in fewer prompts, with high design adherence when working from a reference screenshot, image, or full design system; it scores an Elo of 1588 versus 3.6 Flash's 1538 on Arena.ai's WebDev Arena. In knowledge-dense fields such as finance, law, and biosciences, Google reports better reasoning and accuracy: 34.0% versus 22.0% on the GDP.pdf benchmark for processing complex documents, and 30.4% versus 17.0% on AutomationBench, which tests completion of real-world business workflows.

Google also describes a better developer experience: the model adapts to roadblocks, clarifies intent when needed, follows instructions more faithfully, and puts more effort into multi-step planning and tool calls, which Google says means less manual oversight and fewer retries across engineering workflows. Google adds that early customer feedback describes 3.7 Flash's performance and precision as significantly better than 3.6 Flash's, achieved at a low cost. The introductory price is $0.75 per million input tokens and $3.75 per million output tokens, which Google describes as half of 3.6 Flash's original per-token cost, holding through the end of the year.

The same day, Gemini 3.7 Flash also started powering Gemini Spark, the personal AI agent Google introduced at I/O that runs 24/7 and takes action on a user's behalf under their direction. Google says the update makes Spark more efficient for knowledge work through improved tool use for Google Workspace apps, delivering improved accuracy and output quality for complex, multi-skill workflows. With 3.7 Flash, Google says, Spark can consolidate files, draft emails, and update status documents more efficiently. Spark is available to Google AI Pro and Ultra subscribers in more than 160 countries.

Google also says the model ships with updated safety safeguards against misuse in the CBRN (chemical, biological, radiological, nuclear) and cyber-offense domains, while keeping beneficial uses open. Developers can reach 3.7 Flash through Google Antigravity or the Gemini API via Google AI Studio and Android Studio; enterprises get it through the Gemini Enterprise Agent Platform and the Gemini Enterprise app; individual subscribers get it through Spark in the Gemini app.

Key facts

  • Google released Gemini 3.7 Flash just three weeks after Gemini 3.6 Flash, at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens through the end of the year, half of 3.6 Flash's original per-token cost.
  • On coding, 3.7 Flash beats 3.6 Flash 43.6% to 34.4% on the FrontierCode 1.1 Main benchmark and 65.3% to 49.0% on DeepSWE v1.1.
  • On web development and knowledge work, it also leads: an Elo of 1588 versus 1538 on Arena.ai's WebDev Arena, 34.0% versus 22.0% on the GDP.pdf document-processing benchmark, and 30.4% versus 17.0% on AutomationBench.
  • Gemini Spark, Google's 24/7 personal AI agent for Google AI Pro and Ultra subscribers in more than 160 countries, switched to running on 3.7 Flash the same day.
  • The model ships with updated safety safeguards against misuse in the CBRN and cyber-offense domains, while Google says it keeps beneficial uses open.

Why it matters

Google shipped Gemini 3.7 Flash just three weeks after Gemini 3.6 Flash. Google calls it the most capable workhorse model yet for coding and agents, and pairs the performance gains with a price cut: the introductory per-token price is half of 3.6 Flash's original cost, through the end of the year. Google frames that pairing of stronger benchmark scores and a lower price as what lets developers and customers scale production-ready agents cost effectively.

Who it affects

The update reaches developers building with the Gemini API through Google AI Studio and Android Studio, and those building agent-first workflows in Google Antigravity. Businesses using the Gemini Enterprise Agent Platform or the Gemini Enterprise app get the model for enterprise workflows. Individual subscribers to Google AI Pro or Google AI Ultra get it automatically through Gemini Spark, the 24/7 personal AI agent available in more than 160 countries, which switched to running on 3.7 Flash the same day. Google specifically calls out finance, law, and biosciences as knowledge-dense fields where it says the model shows improved reasoning and accuracy.

How to use it

Gemini 3.7 Flash is priced, through the end of the year, at $0.75 per million input tokens and $3.75 per million output tokens, an introductory rate Google describes as half of 3.6 Flash's original per-token cost. Developers can reach it through the Gemini API via Google AI Studio or Android Studio, or build agent-first workflows in Google Antigravity. Enterprises access it through the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Individual users get it automatically through Gemini Spark in the Gemini app, provided they hold a Google AI Pro or Ultra subscription in one of the more than 160 supported countries.

How solid is it

Every claim here (the benchmark scores, the developer-experience description, and the characterization of 'early customer feedback') comes from Google's launch post announcing its own model. No specific customer, researcher, or named spokesperson backs any of it; every statement is made in the company's collective voice, and no individual author or executive is named anywhere in the announcement. The benchmarks (FrontierCode 1.1 Main, DeepSWE v1.1, Arena.ai's WebDev Arena, GDP.pdf, and AutomationBench) are at least named and scored rather than left as an unquantified claim, but the source does not publish methodology, test conditions, or run counts for any of them.

Risks and caveats

The source attaches a footnote marker to 'introductory price,' but the footnote's own content is not present in the captured text, so the stated pricing may carry conditions not shown here. The source also does not say what the price becomes after 'the end of the year': whether it rises, and by how much, is left open. No context-window size, parameter count, multimodality details, or latency figures are given for the model, and 3.6 Flash's original per-token price is never stated as a number, only that 3.7 Flash's rate is half of it. The 'algorithmic innovations' behind the release are not specified.