Gemini 3.7 Flash Operational

Gemini 3.7 Flash is the next-generation flagship AI model from Google DeepMind. Designed specifically for coding and agent workflows, it delivers major gains across software engineering, web development, and enterprise automation benchmarks, surpassing Claude Sonnet 5 and GPT-5.6 Terra on some metrics. Its launch price is only half the original price of 3.6 Flash, positioning it as a “value-for-money powerhouse” for high-frequency agents and enterprise workflows.

Software engineering: Generates high-quality production-grade code, with major improvements on the FrontierCode and DeepSWE benchmarks, significantly enhancing first-pass accuracy and debugging capabilities.
Web development: Generates high-fidelity interactive websites from screenshots, images, or a complete Design System in one step, leading competitors in Code Arena Elo scores.
Agent automation: Executes complex, multistep enterprise workflows, automatically planning, calling tools, and adapting when obstacles arise, with major gains on AutomationBench.
Knowledge-intensive work: Handles complex document reasoning in fields such as finance, law, and life sciences, with significantly improved understanding of long documents such as GDP.pdf.
Multimodal creation: Combined with Nano Banana, transforms text prompts into playable 3D games in real time, or converts static PDFs into dynamic webpages with interactive charts.
Gemini Spark integration: Serves as the backend model for a 24/7 personal agent, powering automated operations in Google Workspace such as file organization, email drafting, and status updates.

Algorithmic iteration and rapid evolution: The model uses post-training optimization and targeted capability enhancement for coding and agent scenarios rather than relying solely on expanded pretraining scale. This allows it to maintain the Flash series’ speed advantage while achieving a major leap in software engineering and knowledge-work quality.
Multistep planning and rigorous execution: The model includes a more proactive reasoning-resource allocation mechanism. When facing complex tasks, it dedicates more computational resources to multistep planning and tool calls, while adapting its strategy and clarifying intent when obstacles arise during execution. This “more rigorous execution” greatly reduces the need for human supervision and repeated retries, serving as the core technical foundation for its advances in agent and enterprise automation capabilities.
Native unified multimodal understanding: 3.7 Flash provides unified understanding across text, audio, images, code, and video. This enables it to combine visual inputs such as screenshots and design drafts with language instructions in scenarios including 3D game generation, closed-loop robot training, and intelligent document conversion, supporting end-to-end generation from static PDFs to interactive webpages and from text descriptions to runnable 3D assets.
Agent workflows and sub-agent orchestration: Optimized at the foundational level for high-frequency agent scenarios, the model supports the main model in coordinating multiple sub-agents to complete tasks through improved instruction-following accuracy and obstacle adaptability. Its orchestration capabilities enable parallel processing of layout generation, interaction-logic coding, and visual-component rendering in scenarios such as web development, making it enterprise automation infrastructure suitable for long-running, large-scale usage.

Developer access: Developers can call the model directly through the Gemini API or Google AI Studio.
Quick start: In Google AI Studio, users can create a project at no cost, select the gemini-3.7-flash model, and quickly test its text, code, and multimodal capabilities.
Mobile integration: Android Studio has integrated the model, enabling developers to access code assistance and generation during mobile app development.
Agent orchestration: Through Google Antigravity, an agent-first development environment, users can orchestrate sub-agents to build complex automated workflows.
Enterprise deployment: Enterprise users can customize business processes and tool integrations through the Gemini Enterprise Agent Platform, enabling high-frequency usage at scale.
Enterprise use: Employees can also access the built-in 3.7 Flash capabilities through the unified Gemini Enterprise App interface to process documents and perform automation tasks.

Exceptional iteration speed: The Flash series evolved from 3.5 to 3.7 in just 3 months, while the transition from 3.6 to 3.7 was compressed to just 3 weeks. Google’s response to developer feedback and market competition is unmatched among leading models.
Major coding gains: FrontierCode 1.1 Main increased from 34.4% to 43.6%, while DeepSWE v1.1 rose from 49% to 65.3%. First-pass code accuracy and long-cycle software engineering capabilities have both improved dramatically, delivering code quality suitable for a production-grade primary model.
Leading agent and enterprise automation: With an AutomationBench score of 30.4%, the model nearly doubled its predecessor’s result while pulling far ahead of Claude Sonnet 5 (10.7%) and GPT-5.6 Terra (23.6%), making it the strongest model currently available for enterprise workflow automation.
Surpassing leading competitors on select benchmarks: On the two key coding benchmarks FrontierCode 1.1 (43.6%) and WebDev Code Arena (1588 Elo), 3.7 Flash has overtaken Claude Sonnet 5 and GPT-5.6 Terra, breaking the stereotype that “Flash is only for lightweight assistance.”
Exceptional value barrier: Promotional pricing is $0.75 per million input Tokens and $3.75 per million output Tokens, just half the original price of 3.6 Flash and one-third to one-quarter the price of Claude ($2/$10) and GPT ($2/$12). In high-frequency agent and enterprise workflows, this cost advantage scales exponentially with usage.
Native end-to-end multimodal creation: Supports direct generation of playable 3D games from text, one-step conversion of screenshots or Design Systems into interactive webpages, and automatic transformation of static PDFs into dynamic data stories with real-time charts, providing a complete creative pipeline without stitching together multiple models.

Official website: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/

Intelligent coding assistant: Automatically generates production-grade code, debugs complex bugs, and completes long-cycle software engineering tasks, with a major improvement in first-pass accuracy over the previous generation.
Interactive web development: Generates high-fidelity interactive webpages from screenshots or a Design System in one step, with support for multi-agent collaboration and parallax effects.
Enterprise workflow automation: Powers RPA and business systems to automatically complete high-frequency repetitive tasks such as cross-platform data entry, approval routing, and report generation.
3D games and content generation: Combined with Nano Banana, transforms text descriptions into playable 3D games in real time, dynamically generating characters, props, and textures.
Intelligent document conversion: Automatically transforms static PDF annual reports and research reports into dynamic webpages or data stories with real-time charts, data summaries, and interactive features.