Google DeepMind rolled out Gemini 3.7 Flash on August 13, a refresh of its workhorse model that arrived just three weeks after Gemini 3.6 Flash. The company calls the new release its most intelligent option yet for coding and agents, and it pairs the upgrade with a temporary 50 percent price cut on API tokens.
Until December 31, developers pay $0.75 per million input tokens and $3.75 per million output tokens. Standard pricing doubles on January 1, 2027, to $1.50 and $7.50, so the discount gives teams a runway to test whether fewer retries offset the eventual increase.
Google’s benchmark tables show real gains. Gemini 3.7 Flash scores 43.6 percent on FrontierCode 1.1 Main, up from 34.4 percent for 3.6 Flash and ahead of the 42.7 percent Google lists for Claude Sonnet 5 and 41.3 percent for GPT-5.6 Terra. It reaches 65.3 percent on DeepSWE v1.1, trails Terra on Terminal-bench 2.1, and jumps to 30.4 percent on AutomationBench from 17.0 percent. The model also improves on complex PDF comprehension and web layout generation.
The pitch is disciplined execution. Google says 3.7 Flash thinks more diligently, plans multi-step work, and needs fewer human interventions, which matters for enterprise agents that chain tool calls across documents and systems. The model is live in Gemini Spark for AI Pro and Ultra subscribers and inside the Gemini Enterprise Agent Platform.
One gap remains: Gemini 3.5 Pro still has no release date after months of partner testing, leaving the Flash line to carry Google’s momentum in the agent race.
