Gemini 3.7 Flash Tested: Major Boosts in Coding and Reasoning at Half the Cost

petrusenko_max · x · 2026-08-14

The author compares the performance of Gemini 3.7 Flash against version 3.6. Data shows significant improvements in coding accuracy (FrontierCode benchmark rose from 34.4% to 43.6%) and knowledge work reasoning (GDP.pdf benchmark increased from 22% to 34%).

Additionally, the new model generates more complete web applications with fewer prompts, while cutting the cost per million tokens by 50%, achieving a dual breakthrough in performance and cost-efficiency.

Original post →

More from Models

Models channel →