Gemini 3.7 Flash: Google deprioritizes cutting-edge models to compete with Chinese AIs on price

Gemini 3.7 Flash arrived Thursday, twenty-three days after Gemini 3.6 Flash. Such a pace is no longer a matter of a product schedule but of a machine gun, and Google is embracing it. The Flash range is for fast, inexpensive models, designed for high-volume use and automated agents.What is still conspicuously missing, however, is Gemini 3.5 Pro, which was promised at the May developer conference, tested with partners in July, and has been nowhere to be seen since.

Plenty of Flash, still no Pro

On paper, the new version makes significant strides in programming. Google claims a jump from 49% to 65.3% on the DeepSWE test, and from 34.4% to 43.6% on another coding exercise. The task-automation score, meanwhile, nearly doubles. These are in-house measurements, with no published independent validation, and therefore require appropriate caveats.

The previous week, the DeepMind division underwent a complete reshuffle. Its longtime head stepped aside in favor of his deputy, while the two technical leaders behind Gemini walked out to found their own company. Sergey Brin, Google’s co-founder, has spent months pushing the teams to devote themselves entirely to Gemini. When a group ships an intermediate model every three weeks while leaving its flagship model in the garage, that is generally not a sign of a comfortable lead.

The price war has already been lost before it even begins

Hence the real announcement, the one buried in the fine print. Gemini 3.7 Flash is priced at $0.75 per million input text tokens and $3.75 for output, half the launch price of the previous model. In reality, this is a promotion valid until the end of the year, after which the full price returns.Google has therefore chosen its battleground: rather than chasing the best model in the world, selling computing power in bulk at the lowest possible price.

Except that this territory is already firmly occupied. DeepSeek currently charges around $0.14 per million input tokens for its fast version, five times less than Google. On top of that, the company releases its models under open licenses, allowing any company to install them on its own servers without paying anyone. Qwen, from Alibaba, plays the same tune with a permissive license and hundreds of variants available for download. GLM, Kimi, and MiniMax round out a Chinese offering that has turned the price war into an industrial strategy rather than a back-to-school promotion.

What remains is the fate reserved for Europeans, and it is anything but trivial. The new model powers Spark, Google’s personal agent, which is designed to perform tasks on your behalf and is available exclusively to paying subscribers, having been rolled out in more than 160 countries. The European Economic Area, Switzerland, and the United Kingdom are excluded, with no official explanation whatsoever, just as Apple had postponed its next-generation Siri in Europe. French developers, meanwhile, can indeed use Gemini 3.7 Flash in dedicated tools. For everyone else, this new model is little more than news they can watch go by.

👉🏻 Follow the tech news in real time: add 01net to your feeds on Google, and subscribe to our WhatsApp channel.

Source: Gemini