Introducing Gemini 3.7 Flash

Details

Google DeepMind officially launched Gemini 3.7 Flash just three weeks after 3.6 Flash, citing "developer feedback and algorithmic innovations" — not a new pretraining run — as the driver. The rapid iteration cycle itself is a signal: Google says the innovations will carry forward to future models, suggesting this is a repeatable optimization pipeline rather than a one-off improvement. Benchmark gains concentrate in three areas: software engineering (FrontierCode 1.1 Main: 34.4% to 43.6%; DeepSWE v1.1: 49.0% to 65.3%), web development (WebDev Arena Elo: 1538 to 1588), and knowledge-dense document work (GDP.pdf: 22.0% to 34.0%; AutomationBench: 17.0% to 30.4%). The model is positioned specifically for legal, financial, and biosciences document pipelines based on AutomationBench results. Pricing: $0.75/1M input tokens, $3.75/1M output tokens, guaranteed through end of year. This is 50% below original 3.6 Flash pricing and roughly a third the cost of Claude Sonnet 5. The blog demonstrates four concrete use cases: real-time 3D game generation, single-shot interactive landing page creation via sub-agent orchestration, robotics policy training using a 3-agent graph loop, and PDF-to-interactive-data-story conversion. The model is available through the Gemini API, Google AI Studio, Antigravity, Android Studio, Gemini Enterprise Agent Platform, and Gemini Enterprise app. Consumer access via Gemini Spark on AI Pro and Ultra plans. No self-hosting option.