Google launches Gemini 3.7 Flash; API costs cut by 50%, workflow and coding capabilities enhanced

Summary

Google released Gemini 3.7 Flash just three weeks after version 3.6. The company is cutting API prices per token by 50% through the end of 2026. Additionally, it claims that the model achieved the highest scores in Zapier’s AutomationBench benchmark, outperforming Claude Sonnet 5 and GPT-5.6 Terra. These gains in coding, multi-step planning, and document reasoning stem from algorithmic optimization within the same architectural generation, rather than a new pre-training cycle.

Why it matters

For enterprise teams building agentic workflows, a cheaper and more disciplined Flash-class model directly reduces the cost of long-context and multi-step tool-calling loops that dominate real-world automation. The 50% price advantage, valid until December 2026, creates a concrete adoption deadline. On the other hand, leadership changes at DeepMind and the four delayed launch dates for the flagship Pro model add uncertainty to Google’s broader AI roadmap.

Highlights

• FrontierCode 1.1 and DeepSWE v1.1 scores rose from 34.4% to 43.6% and from 49.0% to 65.3%, respectively, within just three weeks. All these improvements occurred within the same architecture.
• The 50% discount per token is a launch promotion that expires on December 31, 2026. The “final” price will take effect again on January 1, 2027.
• The announcement came eight days after DeepMind lost both of its original Gemini co-technical leads. This departure caused Alphabet shares to drop by approximately 4%.

Critical take

The claim that it is “outperforming competitors” relies on a single benchmark published by Zapier. Absolute scores remain modest (43.6% on FrontierCode, 34% on GDP.pdf), indicating that significant room for improvement still exists. When a temporary price cut is paired with an incremental update from the same generation, and the flagship Pro model has missed four launch dates, it is enough to interpret this move as a user retention strategy rather than a true generational leap.