Google unveils cheaper, smarter Gemini 3.7 Flash model for coding
What's the story
Google has launched its latest artificial intelligence (AI) model, Gemini 3.7 Flash, just three weeks after the release of 3.6 Flash. The new model is designed to enhance coding, web development, and knowledge tasks by leveraging developer feedback and smarter algorithms. Senior Director Tulsee Doshi said that Gemini 3.7 Flash is significantly better at coding than its predecessor.
Performance boost
Significant improvements in coding and web development
The new Gemini model shows major improvements in its performance metrics.
The DeepSWE v1.1 score improved from 49% to 65.3%, while the FrontierCode 1.1 Main performance jumped from 34.4% to 43.6%.
These numbers indicate a significant leap in the capabilities of the latest AI model, especially when it comes to coding tasks and web development projects.
Knowledge processing
Better at handling complex information
Along with coding and web development, Gemini 3.7 Flash also excels at knowledge-based tasks.
The model's GDP.pdf benchmark accuracy has improved from 22% to 34%. It also scored better on AutomationBench, with a score of 30.4% as compared to the previous version's score of just 17%.
These improvements show how advanced the latest AI model is at handling complex information and executing common business workflows.
Cost details
Pricing and availability of the new model
Gemini 3.7 Flash is available at a cost of $0.75 per million input tokens and $3.75 per million output tokens. This is half of what 3.6 Flash model costs.
The model can be accessed via Spark (AI Pro and Ultra subscription required), Google Antigravity, or AI Studio.
However, its availability for individual users is limited as it only powers the Gemini Spark agent in the Gemini app for those with an AI Pro or Ultra subscription.