Google has announced the Gemini 3.7 Flash model, released just three weeks after its predecessor, Gemini 3.6 Flash. The new model is positioned as Google's most intelligent workhorse model yet for coding and agents, according to the company.
Benchmark improvements
Gemini 3.7 Flash shows significant gains across several benchmarks compared to Gemini 3.6 Flash. On the DeepSWE v1.1 benchmark, the model scores 65.3%, though one source reports the score as 65.5%. The previous model scored 49.0% on the same test. On FrontierCode 1.1 Main, the model scores 43.6%, up from 34.4%.
Gemini 3.7 Flash's DeepSWE v1.1 benchmark score, compared to 49.0% for the previous model.
The model also improves in web development and document processing. Its WebDev Arena Elo score is 1588, up from 1538. On the GDP.pdf benchmark, which measures how well a model processes complex documents, the model scores 34%, compared to 22% for Gemini 3.6 Flash. On AutomationBench, which tests execution of common business workflows, the score is 30.4%, up from 17%.
In the Agent Arena multi-modal ranking, Gemini 3.7 Flash ranks 20th. Sources differ on the previous model's position: one report places Gemini 3.6 Flash at 30th, while another mentions both 30th and 35th in different sections.
Pricing and availability
Until the end of 2026, Gemini 3.7 Flash has an introductory price of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens. This is half the price of the previous model at launch. Starting January 1, 2027, the price will increase to $1.50 per 1 million input tokens and $7.50 per 1 million output tokens.
Gemini 3.7 Flash is rolling out to Spark in the Gemini app for Google AI Pro and Ultra subscribers. It is also available in Google Antigravity, AI Studio, Android Studio, Gemini Enterprise Agent Platform, and the Gemini Enterprise app. The model is additionally available in the Gemini API.
Gemini 3.7 Flash is noticeably better at coding than the previous Flash release.
The update includes revised safeguards against misuse in chemical, biological, radiological, and nuclear (CBRN) and cyber offense domains.
