Gemini 3.7 Flash Launches with Halved Pricing and Multiple Benchmark Improvements

On August 13, 2026, Google launched the Gemini 3.7 Flash model, focused on coding and agentic work, priced at $0.75 per million input tokens and $3.75 per million output tokens until the end of the year—half the price of the original Gemini 3.6 Flash. The model shows improved performance across multiple benchmarks and is now available in over 160 countries.

The Facts

This release comes just three weeks after Gemini 3.6 Flash. The official blog disclosed that Gemini 3.7 Flash is an algorithmic improvement rather than a newly pretrained model, supporting a 1M-token context window and up to 64K output tokens. The introductory price is set directly at half the previous generation's original price, with the promotional period running until December 31, 2026. The model is now available on the Gemini API, AI Studio, Antigravity, and Gemini Enterprise platforms, and has been integrated into the Spark feature of the Gemini App to handle multi-step tasks across Gmail, Google Calendar, and Google Docs.

Mechanism Breakdown

The price cut stems from lower production costs and algorithmic optimization. Official materials show that 3.7 Flash delivers substantial improvements in software engineering, knowledge work, and web development workflows, including better debugging, problem-solving, and first-pass code accuracy. It reaches 43.6% on FrontierCode 1.1 Main (versus 34.4% for 3.6 Flash) and 65.3% on DeepSWE v1.1 (versus 49.0%). In web development, it scores 1588 on the Arena.ai WebDev Arena Elo (versus 1538). In knowledge-intensive domains such as finance, law, and biosciences, it scores 34.0% on the GDP.pdf benchmark (versus 22.0%) and 30.4% on AutomationBench (versus 17.0%). These improvements come directly from developer feedback and algorithmic iteration.

Industry Impact

For developers, lower token costs for the same tasks directly reduce spending on everyday coding and agentic workloads. Enterprise users can redirect the saved budget toward multi-step workflow automation, such as converting PDFs into interactive data stories or generating playable 3D games. In the competitive landscape, this move strengthens multi-vendor selection, and other models will need to adjust their price-performance ratios or response speeds in specific scenarios. The integration of upstream and downstream toolchains such as AI Studio and Gemini Enterprise lowers the barrier to adopting the new model.

Comparison and Precedents

Compared with Gemini 3.6 Flash released three weeks ago, 3.7 Flash shows consistent improvements on the same benchmarks, with the price cut in half. Artificial Analysis data shows an intelligence index score of 56, up 4 points from the previous generation, placing it on the Pareto frontier of intelligence versus single-task time. Media reports such as MarkTechPost and The Decoder both note that this is an algorithmic improvement over the previous model rather than a new training run, making the pricing strategy change the core differentiator of this release.

Strategic Assessment

Based on available benchmark data and pricing information, Gemini 3.7 Flash is most likely to be adopted as a primary model by more developers during the year-end promotional period. After prices revert to $1.50/$7.50 per million tokens on January 1, 2027, changes in actual usage, along with the completion rate of multi-step agentic tasks in real Google Workspace environments, can serve as signals for whether the model maintains its current cost-performance advantage. These conclusions are anchored in publicly available benchmark comparisons and pricing facts.

When selecting a model, developers can prioritize testing 3.7 Flash's first-pass accuracy and tool-calling stability in coding debugging and document processing scenarios. Enterprises can use the promotional window to evaluate its actual token consumption in automated workflows before deciding on long-term deployment ratios. All conclusions are anchored to officially disclosed benchmark figures and pricing terms.