Google’s cheap model is now two versions ahead of its flagship

Google has released Gemini 3.7 Flash with sharp gains on coding benchmarks and introductory pricing of $0.75 per million input tokens. Gemini 3.5 Pro remains months behind schedule, and Google will not say whether it is still coming.

Google has released Gemini 3.7 Flash, and still will not say when its flagship model is coming. The workhorse model posts large gains on coding, which is the ground Google has been losing. Gemini 3.5 Pro, the model this one is meant to sit beneath, remains months behind schedule.

The numbers are real. On FrontierCode 1.1 the new model scores 43.6% against 34.4% for its predecessor, on DeepSWE v1.1 it reaches 65.3% against 49.0%, and on AutomationBench it more than doubles, from 17.0% to 30.4%.

Google is pricing it to be used. Input costs $0.75 per million tokens and output $3.75 until the end of December, after which both double. It also powers Gemini Spark for AI Pro and Ultra subscribers across more than 160 countries.

The safety work is pitched at the frontier categories. Google says it added safeguards against cyber offence and chemical, biological, radiological and nuclear misuse, in line with its Frontier Safety commitments.

What stands out is the cadence. This lands three weeks after the last Flash release, and Google put out three Flash models in July alone, including a security-tuned variant.

The Pro line has not moved in that time. Google’s next Pro model is months behind schedule after its coding performance fell short of internal targets, which is awkward given coding is where the money is.

It may not arrive at all. Axios reports Google could skip 3.5 Pro entirely and go straight to Gemini 4 Pro, and the company declined to discuss the model’s fate.

Sundar Pichai told investors in July that Google intends to ship models faster and is already spending heavily on compute to train Gemini 4. Shipping the cheap tier every three weeks is one way to look faster while the expensive one stalls.

There is a cost inside the building. The delay has been hard on DeepMind morale, at a moment when rival labs are hiring.