Introducing Gemini 3.7 Flash
2026-08-20 · Google DeepMind
Introducing Gemini 3.7 Flash
On August 13, 2026, Google announced Gemini 3.7 Flash — its most intelligent workhorse model yet, purpose-built for coding and agentic workflows.
The release arrives just three weeks after Gemini 3.6 Flash, driven by direct developer feedback and recent algorithmic innovations that the team plans to carry forward into future models.
Major Intelligence Gains for Complex Workflows
Gemini 3.7 Flash demonstrates substantial improvements over its predecessor across several domains:
Software Engineering
- Stronger performance in debugging and issue resolution
- Higher first-pass code accuracy
- Better generation of production-ready code
Benchmark Results
- FrontierCode 1.1 Main: 43.6% (vs 34.4% for 3.6 Flash)
- DeepSWE v1.1: 65.3% (vs 49.0% for 3.6 Flash)
Web Development
- Generates more functional layouts and feature-complete applications in fewer prompts
- High design adherence and visual parity when given reference inputs (screenshots, images, or design systems)
- WebDev Arena Elo score improved to 1588 (vs 1538 for 3.6 Flash)
Knowledge-Intensive Domains
- Enhanced reasoning and accuracy in finance, law, and biosciences
- GDP.pdf benchmark: 34.0% (vs 22.0%)
- AutomationBench: 30.4% (vs 17.0%)
Real-world demonstrations showcased by the team include generating a fully playable 3D game from a text prompt (combined with Nano Banana for real-time character, item, and texture creation), producing stunning interactive landing pages with parallax components orchestrated via sub-agents and Gemini Omni, training robotics models using multimodal understanding in a three-agent graph loop, and transforming static PDF annual reports into interactive data stories with live charts and aggregated insights.
Improved Developer Experience and Competitive Pricing
Developers report a noticeably better experience with 3.7 Flash. The model is more adept at navigating roadblocks, clarifying ambiguous intent, and following instructions with higher fidelity. It engages in more diligent multi-step planning and tool calling, resulting in more disciplined execution that reduces the need for manual oversight and repeated retries.
Pricing
- Introductory rate (until December 31, 2026): $0.75 per million input tokens and $3.75 per million output tokens
- Standard rate (from January 1, 2027): $1.50 / $7.50 per million tokens
This pricing, combined with the performance uplift, enables cost-effective scaling of production-grade agents.
Early customer feedback highlights significantly better results than 3.6 Flash at a competitive cost.
Powering Gemini Spark
Starting today, Gemini Spark — the 24/7 personal AI agent available to Google AI Pro and Ultra subscribers in over 160 countries — is powered by Gemini 3.7 Flash. Originally launched at I/O, Spark acts under user direction to complete tasks across Google Workspace.
The upgrade improves tool-use efficiency and output quality for complex, multi-skill workflows such as consolidating files, drafting emails, and updating status documents.
Built with Safety in Mind
Gemini 3.7 Flash ships with updated Frontier Safety safeguards, particularly strengthened protections against misuse in Chemical, Biological, Radiological, and Nuclear (CBRN) domains and cyber offense, while preserving beneficial use cases in line with Google’s bioresilience and cyber safety programs.
For detailed information, refer to the 3.7 Flash model card.
Availability
- Developers: Start building in Google Antigravity, Gemini API, Google AI Studio, or Android Studio.
- Enterprises: Available via Gemini Enterprise Agent Platform and the Gemini Enterprise app.
- Individuals: Accessible through Spark in the Gemini app for Google AI Pro and Ultra subscribers.
Gemini 3.7 Flash represents a rapid iteration that meaningfully raises the intelligence bar for efficient, production-ready AI agents.