Gemini 3.7 Flash is $0.75/$3.75 until Dec 31 — same sticker as 3.6, 35% cheaper on one agent's bill

Google shipped Gemini 3.7 Flash on August 13, 2026, three weeks after 3.6 Flash. The model id is gemini-3.7-flash. Intro pricing is $0.75/$3.75 per million tokens through December 31, then $1.50/$7.50. Official DeepSWE v1.1 is 65.3% versus 48.6% for 3.6. One early customer measured a 35% cheaper agent bill at the same list price.

Google released Gemini 3.7 Flash on August 13, 2026, three weeks after Gemini 3.6 Flash. (Source: Google blog, 2026-08-13)

Key facts:

The API model id is gemini-3.7-flash. It is generally available. (Source: Gemini API changelog, 2026-08-13)
Input context is 1 million tokens. Maximum output is 64,000 tokens. Inputs are text, image, video, audio and PDF. Output is text. (Source: DeepMind Gemini 3.7 Flash page)
Introductory price is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. On January 1, 2027 the rate becomes $1.50 / $7.50. The same intro footnote now applies to 3.6 Flash as well.
On DeepMind’s comparison table, DeepSWE v1.1 is 65.3% versus 48.6% for 3.6 Flash. FrontierCode 1.1 Main is 43.6% versus 34.4%. Code Arena Elo is 1588 versus 1538.
Terminal-bench 3.0 is 14.9%, up from 5.4% on 3.6, still far behind Terminal-bench 2.1 at 85.8%.
It is live in the Gemini API, Google AI Studio, Android Studio, Google Antigravity, Gemini Enterprise, and Gemini Spark for Google AI Pro and Ultra subscribers.

Official DeepMind comparison table: Gemini 3.7 Flash at $0.75/$3.75 intro pricing, DeepSWE v1.1 65.3%, FrontierCode 43.6%, Code Arena Elo 1588, Terminal-bench 2.1 85.8%, Terminal-bench 3.0 14.9%, versus Gemini 3.6 Flash, Claude Sonnet 5, GPT-5.6 Terra and Muse Spark 1.2. — DeepMind's own card. The asterisk on both Flash columns is the same Dec 31 intro window. (Table: Google DeepMind, 2026-08-13)

What this means if you’re building with Gemini Flash

1. The list price is not a 3.7 exclusive. Google’s launch post calls $0.75 / $3.75 “half the original 3.6 Flash cost.” The original 3.6 launch rate was $1.50 / $7.50. DeepMind now prints $0.75 / $3.75 on both 3.6 and 3.7, and the same Jan 1, 2027 snap-back. If your finance model still has 3.6 at $1.50, update the row. If you treat the intro as permanent, December 31 is the cliff.

2. Same sticker can still cut the invoice. Browser Use CTO Gregor Zunic, quoted on the DeepMind page: the Gemini 3.7 Flash agent was 35% cheaper than 3.6 Flash, with an +8% prompt-cache hit rate and fewer tool errors. That is a customer measurement, not a Google benchmark. It is the reason to A/B the model id even though the price card looks identical.

Official DeepSWE v1.1 cost-versus-score plot highlighting gemini-3.7-flash near 65% at a lower average cost per task than gemini-3.6-flash, with Claude Opus 5 and Fable 5 higher on the score axis and to the left on cost. — Google's DeepSWE v1.1 cost plot. 3.7 sits above 3.6 on the same chart, closer to the "most efficient" edge. (Chart: Google blog / Datacurve AI)

3. Do not swap a terminal agent on the 85.8% headline. Terminal-bench 2.1 is 85.8%. Terminal-bench 3.0 is 14.9%. GPT-5.6 Terra is 20.8% on v3.0. If the agent lives in a shell, run your own eval. The two Terminal-bench numbers are both official and both true.

4. It is not uniformly better. CharXiv Reasoning without tools is 84.5% for 3.7 versus 85.2% for 3.6. With tools: 88.7% versus 89.4%. GDP.pdf (34.0% vs 22.0%) and AutomationBench (30.4% vs 17.0%) are the large knowledge-work jumps. OSWorld-2.0 goes 33.8% → 47.9%.

5. The migration is a model-id change. Point the same Gemini API client at gemini-3.7-flash. Our Gemini 3.6 Flash guide still covers the Flash-versus-Lite routing decision; Lite remains the high-volume cheap tier. Spark users on Gemini Spark pick up 3.7 automatically. Antigravity users see it in Antigravity 2.0.

Sources: Google — Introducing Gemini 3.7 Flash · DeepMind — Gemini 3.7 Flash · Gemini API changelog · 9to5Google