Glonce

Google releases Gemini 3.7 Flash, halves the price of 3.6

modelsThe Glonce editorsGoogle DeepMind🔥71Glonce rating71Impact70Novelty50Relevance100Surprise38Players95Credibility90
Google releases Gemini 3.7 Flash, halves the price of 3.6

Google DeepMind introduced Gemini 3.7 Flash today, calling it its most intelligent workhorse model yet for coding and agents. The release comes just three weeks after Gemini 3.6 Flash, and the company describes it as a direct result of developer feedback and algorithmic innovations it plans to bring to future models. 3.7 Flash ships at an introductory price of half the original 3.6 Flash cost per million tokens: $0.75 per million input tokens and $3.75 per million output tokens, available at that price through the end of the year.

On coding benchmarks, 3.7 Flash scores 43.6% versus 34.4% for 3.6 Flash on FrontierCode 1.1 Main, and 65.3% versus 49.0% on DeepSWE v1.1. On Arena.ai's WebDev Arena it posts an Elo score of 1588 versus 1538. Google DeepMind says 3.7 Flash shows strong gains over 3.6 Flash in tasks like debugging and issue resolution, achieves higher first-pass code accuracy, and generates more functional layouts and feature-complete web apps in fewer prompts, including high design adherence when working from a reference screenshot, image, or full design system.

For knowledge work, 3.7 Flash scores 34.0% versus 22.0% on the GDP.pdf benchmark, an eval for processing complex documents, and 30.4% versus 17.0% on AutomationBench, which tests completion of real-world business workflows. Google DeepMind says the model delivers improved reasoning and accuracy for knowledge-dense fields such as finance, law, and biosciences, and offers a better developer experience: it adapts to roadblocks, clarifies intent when needed, follows instructions with greater fidelity, and puts more effort into multi-step planning and tool calls, meaning less manual oversight and fewer retries across engineering workflows.

Gemini Spark, the personal AI agent Google launched at I/O and made available to Google AI Pro and Ultra subscribers in over 160 countries, starts using Gemini 3.7 Flash today. The update is meant to make Spark more efficient at knowledge work, with improved tool use for Google Workspace apps and better accuracy on complex, multi-skill workflows such as consolidating files, drafting emails, and updating status documents.

Gemini 3.7 Flash ships with updated safeguards against misuse in the domains of Chemical, Biological, Radiological, and Nuclear (CBRN) threats and cyber offense, part of Google's approach to bioresilience and its cyber program. Developers can reach the model through Google Antigravity, the Gemini API in Google AI Studio, and Android Studio; enterprises through the Gemini Enterprise Agent Platform and the Gemini Enterprise app; and individuals through Spark in the Gemini app.

Key facts

  • Gemini 3.7 Flash launched three weeks after Gemini 3.6 Flash, priced at $0.75/1M input tokens and $3.75/1M output tokens, an introductory rate Google calls half the prior model's cost per million tokens, running through the end of the year.
  • Coding benchmarks versus 3.6 Flash: FrontierCode 1.1 Main 43.6% vs 34.4%, DeepSWE v1.1 65.3% vs 49.0%, WebDev Arena Elo 1588 vs 1538.
  • Knowledge-work benchmarks versus 3.6 Flash: GDP.pdf document-processing eval 34.0% vs 22.0%, AutomationBench 30.4% vs 17.0%.
  • Gemini Spark, available to Google AI Pro and Ultra subscribers in over 160 countries, switched to 3.7 Flash starting today for better Google Workspace tool use.
  • The model ships with updated safeguards against CBRN and cyber-offense misuse.

Why it matters

This is Google's second Flash-tier update in three weeks, following Gemini 3.6 Flash, and Google DeepMind frames it as a fast-cadence example of turning developer feedback into shipped model improvements. The gains concentrate in coding and agentic workflows: higher first-pass code accuracy, better debugging and issue-resolution performance, and more disciplined multi-step planning and tool calling. Google says the result is less manual oversight and fewer retries in engineering workflows, positioning 3.7 Flash as the workhorse tier for production coding agents rather than a flagship reasoning model.

Who it affects

Developers building coding agents and web apps get access to 3.7 Flash through the Gemini API, Google AI Studio, and Android Studio, or through Google Antigravity for agent-first workflows. Enterprises reach it through the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Individual Google AI Pro and Ultra subscribers in over 160 countries get it automatically, since Gemini Spark, the personal AI agent launched at I/O, switched to 3.7 Flash starting today.

How to use it

3.7 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens, an introductory rate the company describes as half the original 3.6 Flash cost per million tokens; it holds through the end of the year. Developers can start building via the Gemini API in Google AI Studio or Android Studio, or through Google Antigravity for agent-first workflows, with a developer guide available for onboarding. For UI generation, the model can work from a reference screenshot, an image, or a full design system.

How solid is it

The improvement claims rest entirely on Google's own benchmark comparisons against its predecessor: FrontierCode 1.1 Main, DeepSWE v1.1, Arena.ai's WebDev Arena Elo rating, the GDP.pdf document-processing eval, and AutomationBench. Every score is presented only against 3.6 Flash; the announcement makes no comparison to any competing model from another AI lab.

Risks and caveats

The source is a company announcement, not an independent review, and every benchmark in it compares 3.7 Flash only against its immediate predecessor. Google states the introductory price applies through the end of the year without giving an exact end date, and does not specify whether input and output prices individually equal half of 3.6 Flash's rates or whether the 'half' figure holds only in aggregate. No architecture, parameter count, or training methodology is disclosed.