Google Gemini 3.7 Flash Launches with Greater Efficiency, No Pro in Sight

Google has recently launched Gemini 3.7 Flash, just three weeks after releasing Gemini 3.6 Flash. This new model emphasizes enhanced software engineering, web development, knowledge work, and agentic tasks, all while preserving the speed characteristic of Google's Flash models. Google credits developer feedback and algorithmic improvements for the enhancements, suggesting the changes may affect future Gemini models.

One of the most notable improvements is in coding tasks, with Gemini 3.7 Flash scoring 65.3% on the DeepSWE v1.1 benchmark for long-horizon software engineering, up from 48.6% in its predecessor. Similarly, FrontierCode 1.1 Main shows progress, increasing from 34.4% to 43.6%. Google claims the model is more adept at managing obstacles, adhering to instructions, and executing multi-step plans with reduced human assistance.

In the realm of web development, Google has enhanced Gemini 3.7 Flash's capabilities. The model can now create more comprehensive and functional applications with fewer prompts, and it adheres more closely to reference designs. The Code Arena web development score improves from 1,538 Elo for Gemini 3.6 Flash to 1,588 Elo. Additionally, Gemini 3.7 Flash can accurately recreate interfaces from screenshots, images, and full design systems.

The model's performance in knowledge-intensive areas like finance, law, and biosciences has also been bolstered. Gemini 3.7 Flash achieves a 34.0% score on the GDP.pdf benchmark for understanding complex PDFs, an increase from 22.0% in the previous version. AutomationBench scores rise from 17.0% to 30.4%, reflecting stronger performance on actual business workflows.

Google has further updated its safety measures to address risks related to chemical, biological, radiological, and nuclear threats, as well as cyber misuse, aligning with the safety protocols applied to its recent Flash models.

To make Gemini 3.7 Flash more accessible, Google has introduced lower introductory pricing. The model is available at $0.75 per million input tokens and $3.75 per million output tokens until December 31, 2026. After this date, pricing will rise to $1.50 and $7.50, respectively. Users can access Gemini 3.7 Flash through various platforms, including the Gemini app, Google AI Studio, Gemini API, Google Antigravity, Gemini Enterprise, and the Gemini Enterprise Agent Platform.