Gemini 3.7 Flash is the fastest model on the market. Its price doubles on January 1

Google launched Gemini 3.7 Flash on August 13, three weeks after Gemini 3.6 Flash. The benchmark motion in that hole is the most important of any latest iteration, and the pricing carries a element most protection has skipped: the quantity on the web page expires.

What Modified In Three Weeks

On FrontierCode 1.1 Foremost, the mannequin scores 43.6 per cent towards 34.4 per cent for its predecessor. On DeepSWE v1.1 it reaches 65.3 per cent towards 49.0 per cent. On AutomationBench, which measures agentic job completion, it strikes from 17 per cent to 30.4 per cent.

Synthetic Evaluation scores it 56 towards 52 for Gemini 3.6 Flash, and ranks it first of 186 fashions measured on output velocity, at 340.1 tokens per second.

A sixteen-point leap on DeepSWE and a close to doubling on AutomationBench in three weeks isn’t the tempo of a mannequin era. It’s the tempo of a post-training run, and it suggests Google is iterating on the identical base far sooner than its launch numbering implies.

Google lists Gemini 3.7 Flash at $0.75 per million enter tokens and $3.75 per million output tokens. Synthetic Evaluation places the blended price at $0.58 per million, half the $1.16 it recorded for Gemini 3.6 Flash.

These are introductory charges. They run till December 31, 2026. On January 1, 2027, they turn into $1.50 and $7.50 — precisely double.

Google has revealed the rise upfront, which is extra disclosure than the business normally provides. However the sensible impact is a mannequin that’s at present the most affordable quick frontier possibility accessible, on phrases that finish on a hard and fast date roughly 4 months out.

Why That Issues Extra Than It Sounds

Mannequin pricing isn’t a retail choice. It’s an enter price that will get constructed into different firms’ merchandise.

A startup pricing its personal service on $0.58 per million tokens is pricing towards a price with an expiry date. Something constructed between now and December that will depend on these economics faces a doubling of its largest variable price on the flip of the yr, at which level the choices are to boost costs, take in the margin, or migrate to a different mannequin and re-run each analysis that justified the unique selection.

Switching prices on this market are actual however not prohibitive, which is exactly the purpose of introductory pricing. It buys integration. By January, the work of embedding the mannequin in a product is already sunk.

The launch lands in a market the place worth and velocity have turn into the contested floor slightly than uncooked functionality.

OpenAI introduced Ultrafast mode for GPT-5.6 Sol on the identical day, operating on Cerebras {hardware} at as much as 750 output tokens per second — sooner than Gemini 3.7 Flash on throughput, however in restricted preview to chose prospects and with no pricing revealed in any respect.

Moonshot’s Kimi K3, launched in late July with open weights, publishes charges of $3 per million uncached enter tokens and $15 per million output, and may be self-hosted by anybody with the {hardware} to run a 2.8-trillion-parameter mannequin.

Towards that, Gemini 3.7 Flash’s proposition is that it’s quick, low cost and customarily accessible as we speak. Two of these three are assured solely till December 31.

The benchmark features are actual and independently measured, and on velocity the mannequin is genuinely first amongst every thing Synthetic Evaluation tracks.

The pricing is a business instrument slightly than an outline of what the mannequin prices to run. Google is shopping for adoption through the window when adoption is most cost-effective to purchase, and it has been unusually simple about when the window closes.

For anybody constructing on it, the quantity to plan round isn’t $0.75. It’s $1.50, and the date is already on the calendar.