8/18/2026, 1:05:10 PM · foundation-models

Google Releases Gemini 3.7 Flash with 43.6% FrontierCode Performance and Introductory Pricing Through End-2026

Google DeepMind's Gemini 3.7 Flash arrives 23 days after its predecessor with sharp coding and automation benchmark gains and a 50% introductory price cut set to expire January 1, 2027.

Google DeepMind on August 13, 2026, released Gemini 3.7 Flash, a generally available large language model (LLM) positioned as the company's primary workhorse for coding, agentic workflows, and knowledge-intensive tasks. <cite index="13-13">The model arrived 23 days after Gemini 3.6 Flash, which shipped in late July 2026.</cite>

Benchmark Results

The headline gain is in production code quality. <cite index="3-2">On FrontierCode 1.1 Main, which measures production code quality, Gemini 3.7 Flash scores 43.6%, up from 34.4% for Gemini 3.6 Flash.</cite> <cite index="5-12">On that same benchmark, Claude Sonnet 5 reached 42.7% and GPT-5.6 Terra reached 41.3%, according to Google's own published comparison table.</cite>

Software engineering over longer task horizons also improved substantially. <cite index="5-14,5-15">DeepSWE v1.1 tests long-horizon software engineering, meaning multi-file tasks that require sustained context over many steps; Gemini 3.7 Flash scores 65.3%, a large jump from 48.6% for Gemini 3.6 Flash, though GPT-5.6 Terra leads that category at 69.6%.</cite>

Gains extended into web development and enterprise automation. <cite index="1-8">Gemini 3.7 Flash received an Elo score of 1,588 on WebDev Arena, compared with 1,538 for its predecessor.</cite> <cite index="3-13">On AutomationBench, which Google describes as measuring enterprise workflow automation, Gemini 3.7 Flash scores 30.4%, up sharply from 17.0% for Gemini 3.6 Flash.</cite> <cite index="3-15">The model also reaches 34.0% on GDP.PDF, an evaluation of complex PDF comprehension, compared with 22.0% for Gemini 3.6 Flash.</cite>

Context retrieval likewise improved. <cite index="4-7">The long-context GDM-MRCR v2 8-needle score climbed from 91.8% to 97.0%.</cite> <cite index="6-7">The model retains a 1,048,576-token input context window and a 65,536-token output limit, both unchanged from Gemini 3.6 Flash, and carries a knowledge cutoff of March 2026.</cite>

Notably, <cite index="6-4">the AutomationBench figure of 30.4% still means the model fails roughly seven in ten multi-step automation tasks</cite>, a caution for teams evaluating production deployment readiness.

Design Priorities

<cite index="3-5,3-6">With Gemini 3.7 Flash, Google's emphasis shifts toward putting sufficient effort into planning while improving the quality of execution. Google DeepMind said in a post accompanying the release that the model shows gains in debugging and issue resolution, generates more functional web layouts and applications with fewer prompts, and improves reasoning and accuracy on real-world business workflows.</cite> <cite index="6-10">The model accepts text, images, video, audio, and PDF as input.</cite>

Pricing and Availability

<cite index="16-2,16-3">Gemini 3.7 Flash is offered with introductory pricing of $0.75 per one million input tokens and $3.75 per one million output tokens through December 31, 2026. Starting January 1, 2027, standard pricing of $1.50 per one million input tokens and $7.50 per one million output tokens will apply.</cite> That structure represents a 50% discount relative to Gemini 3.6 Flash's original launch rates, which are documented on Google's official model card.

<cite index="14-10">Google has also moved Gemini 3.6 Flash onto the same introductory rate, so the two models cost the same today and the upgrade decision turns on capability rather than price.</cite>

<cite index="7-4,7-5">In the Gemini app, Gemini 3.7 Flash is rolling out to Spark for Google AI Pro and Ultra subscribers. The model is also available in Google AI Studio, Android Studio, the Gemini Enterprise Agent Platform, and the Gemini Enterprise app.</cite> <cite index="10-7">Gemini Spark access excludes the European Economic Area, the United Kingdom, Switzerland, and Nigeria at launch.</cite>

Safety

<cite index="1-11,1-12">Google says the release includes updated safeguards against cyber and chemical, biological, radiological, and nuclear misuse. A model card has been published with further information about safety testing and deployment.</cite> <cite index="17-8">Google's model card notes that Gemini 3.7 Flash may exhibit some of the general limitations of foundation models, such as hallucinations.</cite>

The release continues a pattern of rapid iteration in Google's Flash line as competition with other frontier model providers intensifies across both capability and cost dimensions.

Cross-references

Sources

  1. [1]
    Google launches Gemini 3.7 Flash for coding and AI agents | Digital Watch Observatory
  2. [2]
    Google Releases Gemini 3.7 Flash, Competes With GPT 5.6 Terra & Muse Spark 1.2 On Benchmarks
  3. [3]
    Google’s Gemini 3.7 Flash targets coding and agents with a 50% introductory price cut | VentureBeat
  4. [4]
    Google Ships Gemini 3.7 Flash, Coding Bench Jumps to 43.6% | AI Weekly
  5. [5]
    Gemini 3.7 Flash: Features, Benchmarks, and Pricing | DataCamp
  6. [6]
    Google releases Gemini 3.7 Flash for coding and agents
  7. [7]
    Gemini 3.7 Flash launches three weeks after last model, live in Spark
  8. [8]
    Google Launches Gemini 3.7 Flash: Major Leap in Coding and Agent Capabilities, Three-Week Iteration Sets New Speed Record for Low-Cost Models
  9. [9]
    Gemini 3.7 Flash lands with coding gains and undercuts its three-week-old predecessor's price by 50%
  10. [10]
    Google Launches Gemini 3.7 Flash but Its Low Price Has an Expiry Date - Memeburn
  11. [11]
    Gemini 3.7 Flash Launch: Price, Benchmarks, Access
  12. [12]
    Gemini 3.7 Flash Pricing Explained
  13. [13]
    Agent Platform Pricing | Google Cloud
  14. [14]
    Gemini 3.7 Flash - Model Card — Google DeepMind
  15. [15]
    Gemini 3.7 Flash: our most intelligent workhorse model