Key facts
- Google has released Gemini 3.7 Flash, a new AI model.
- This model replaces the Gemini 3.6 Flash version, which was released three weeks ago.
- Gemini 3.7 Flash shows improved performance in coding benchmarks, with scores rising in tests like FrontierCode 1.1 Main and DeepSWE v1.1.
- The model also demonstrates enhanced capabilities in processing complex documents, as indicated by the GDP.pdf benchmark.
- Google is positioning Gemini 3.7 Flash as a cost-effective option with a lower introductory price.
Google has introduced its latest AI model, Gemini 3.7 Flash, a mere three weeks after the release of its predecessor, Gemini 3.6 Flash. This rapid update aims to enhance Google's competitive standing in the AI landscape, particularly in areas like coding and agentic performance.
According to Google's Senior Director Tulsee Doshi, Gemini 3.7 Flash demonstrates significant improvements over the previous version. In coding benchmarks, the FrontierCode 1.1 Main test saw an increase from 34.4% to 43.6%, and DeepSWE v1.1 improved from 49% to 65.3%. The model's WebDev Arena score also rose to 1588 from 1538.
Beyond coding, Gemini 3.7 Flash shows enhanced capabilities in processing complex documents, with its GDP.pdf benchmark score increasing to 34% from 22%. Performance in executing business workflows, measured by AutomationBench, also saw a substantial jump from 17% with 3.6 Flash to 30.4% with the new 3.7 Flash model.
Google is also addressing market competition by positioning Gemini 3.7 Flash with a lower introductory price, aiming to counter the cost-effectiveness of rival models. This release comes amidst speculation about Google's pace of AI development, particularly following the delayed launch of the flagship Gemini 3.5 Pro.