Google Rolls Out Gemini 3.7 Flash to Subscribers

Google's model card identifies Gemini 3.7 Flash as the next model in its Gemini 3 family and states that it is based on Gemini 3.6 Flash. It accepts text, image, audio, and video inputs, supports up to 1 million input tokens, and can produce up to 64,000 output tokens. The API documentation lists text as the output modality.

According to the model card, configurable thinking controls let developers trade off quality, cost, and latency. The API documentation supports low, medium, and high thinking configurations, while noting that minimal is unsupported and returns an error.

Google describes the release as an improvement in coding, instruction following, and tool calling. In its launch post, Google reports a 43.6% score for 3.7 Flash versus 34.4% for 3.6 Flash on one production-ready-code evaluation, and 65.3% versus 49.0% on another. It also reports a 34.0% score versus 22.0% on GDP.pdf, a benchmark for processing complex documents, and 30.4% versus 17.0% on an unnamed business-workflow evaluation.

Those figures are vendor-reported benchmark results rather than independent measurements. Teams evaluating agentic coding systems generally need to validate performance against their own repositories, tool permissions, test suites, and failure-handling requirements, because benchmark gains do not establish reliability in a particular deployment.

Google lists introductory API pricing of $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Its launch announcement states that pricing changes on January 1, 2027 to $1.50 per million input tokens and $7.50 per million output tokens, respectively.

The model's stable API identifier is gemini-3.7-flash. Google documents it as generally available in the Gemini API. The published model card also says it is available to downstream providers through an API, subject to the relevant terms of use.

NokiaPowerUser reports that the subscriber release improves multi-step reasoning, Workspace integration, code generation, and document handling, and that Gemini Spark can work across connected Gmail, Calendar, and Docs accounts. Google has separately documented the model's coding and agent features, but the retrieved official materials do not independently detail the reported AI Pro, AI Ultra, or Gemini Spark consumer rollout.

For practitioners, the notable technical combination is a long context window, multimodal inputs, and configurable reasoning in a lower-cost Flash-tier model. Comparable model releases make evaluation design especially important: token-cost comparisons should account for thinking-token use, prompt caching, tool-call volume, and the cost of retries, rather than relying on input and output token prices alone.