Gemini 3.7 Flash Ships in Three Weeks, and Your Price Sheet Expires in December

Google released Gemini 3.7 Flash on August 13, 2026, three weeks after Gemini 3.6 Flash, positioning it for coding, agents, knowledge work, and web development ^[1][2][3]. Two details matter more than the benchmark table: the release interval, and the fact that the introductory API price expires on December 31, 2026 ^[4].

The specifications are the ones you would design an agent around. The model card lists text, image, audio, and video inputs with text output, a 1,048,576-token context window, and a 65,536-token maximum output ^[5]. The stable model ID is gemini-3.7-flash ^[6]. Reasoning is configurable across LOW, MEDIUM, and HIGH thinking levels, with MEDIUM as the default, and Google's enterprise documentation notes that MINIMAL is unsupported and returns an API validation error if you set it explicitly ^[7][8]. Distribution covers the Gemini API, AI Studio, Android Studio, Gemini Enterprise Agent Platform, and Google Antigravity ^[9]. Ars Technica reports it also powers the Gemini Spark agent for AI Pro and Ultra subscribers, while the consumer chatbot stayed on 3.6 Flash at launch ^[10].

Google reports gains over 3.6 Flash on every evaluation it cited: FrontierCode 1.1 Main 43.6 percent against 34.4, DeepSWE v1.1 65.3 percent against 49.0, WebDev Arena Elo 1,588 against 1,538, GDP.pdf 34.0 percent against 22.0, and AutomationBench 30.4 percent against 17.0 ^[11]. These are vendor-reported numbers, not independent comparisons, and Ars Technica confirmed the published figures while questioning whether the differences justified another release after three weeks ^[12][13]. Read the AutomationBench figure the other way and the leading number still leaves 69.6 percent of tasks unfinished ^[14]. Benchmark movement does not establish reliability in tool use, permission handling, recovery from failed steps, or task-specific code review ^[15].

The pricing is where a pinned model ID becomes a budget problem. Introductory rates are $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, moving to $1.50 and $7.50 on January 1, 2027 ^[4][16]. That is a doubling on both sides of the ledger, on a fixed date, with output priced at five times input in both tiers ^[17][18]. The source also describes the introductory rate as half the original 3.6 Flash rate, which means January does not introduce a premium so much as end a discount ^[19][20].

Now combine the two facts. From August 13 to December 31 is roughly 20 weeks; at a three-week cadence that is about six more Flash releases before the price changes ^[21]. Any team that pins gemini-3.7-flash, tunes prompts against its MEDIUM default, and builds a cost model on $0.75 per million input tokens is committing to a configuration that will be several versions stale and twice as expensive by the new year. The workload-level tests you need are the same either way: token spend, latency, and cost per completed task, measured on your own tasks rather than on FrontierCode ^[15].

Watch three things. Whether the next Flash release lands around September 3, 2026, which is what three weeks from August 13 implies ^[22]. Whether the consumer chatbot moves off 3.6 Flash, since that gap suggests internal confidence in 3.7 is uneven ^[10]. And Gemini 3.5 Pro, which Ars Technica notes 3.7 Flash arrived ahead of, and for which Google announced no date in the launch materials ^[23][24].