Google has released Gemini 3.7 Flash, a new iteration of its large language model, designed to replace the previous 3.6 Flash version. The update is intended to enhance coding and agentic capabilities based on core optimizations and developer feedback. According to Senior Director Tulsee Doshi, the model shows significant improvements in coding benchmarks, with scores rising from 34.4 to 43.6 percent on the FrontierCode 1.1 Main test and from 49 to 65.3 percent on the DeepSWE v1.1 test. "Gemini 3.7 Flash is noticeably better at coding than the previous Flash release," Doshi said.
The new model also demonstrates modest improvements in general knowledge tasks, with a WebDev Arena score of 1,588 compared to 1,538 for the previous version. Performance on the GDP.pdf benchmark, which measures document processing, has increased to 34 percent from 22 percent, while the AutomationBench score has risen to 30.4 percent from 17 percent. These results suggest that Gemini 3.7 Flash is slightly better than its predecessor in various benchmarks. However, the rapid release cycle has raised questions about whether the improvements justify a new model just three weeks after the last one.
Google is positioning Gemini 3.7 Flash as a cost-effective alternative, with a price of $0.75 per million input tokens and $3.75 per million output tokens, half of what 3.6 Flash costs. This pricing strategy aims to retain developer interest amid competition from models like OpenAI’s Luna, which is priced at $0.20 per million input tokens and $1.20 per million output tokens. The new model is available through the Gemini API, AI Studio, and Gemini Enterprise, but individual users can only access it via the Gemini Spark agent with an AI Pro or Ultra subscription. The regular chatbot interface continues to use the 3.6 Flash version.
Source: arstechnica
