Gemini 4 Pro Could Be an Absolute Beast — Google’s New Flash Benchmarks Have the AI World Buzzing

Google may be cooking up an AI monster behind closed doors—and Gemini 3.7 Flash just gave us a seriously tantalizing taste of what could be coming.

The tech giant has officially confirmed that it is training Gemini 4, describing the project as its “most ambitious” pre-training run yet. Google has not formally announced a model called “Gemini 4.0 Pro,” but speculation is exploding that the company’s next flagship release could leapfrog the delayed Gemini 3.5 Pro entirely.

And honestly? If the newly released Gemini 3.7 Flash is the appetizer, the main course could be downright terrifying for Google’s AI rivals.

Gemini 3.7 Flash Is Punching Way Above Its Weight

Gemini 3.7 Flash is supposed to be Google’s fast and relatively inexpensive workhorse—not its biggest, most powerful frontier model. However, the model’s official benchmark results show it trading punches with, and sometimes beating, much more expensive flagship systems.

According to Google DeepMind’s official model card, Gemini 3.7 Flash scored 56 on the Artificial Analysis Intelligence Index. That placed it just one point behind GPT-5.6 Terra and Muse Spark 1.2, while edging past Claude Sonnet 5’s score of 55.

The coding numbers are even more eye-popping:

• FrontierCode 1.1: 43.6%, beating Claude Sonnet 5’s 42.7% and GPT-5.6 Terra’s 41.3%.
• DeepSWE v1.1: 65.3%, compared with 48.6% for Gemini 3.6 Flash and 53.8% for Claude Sonnet 5.
• Code Arena: An Elo rating of 1,588, topping every comparison model listed by Google.
• Terminal-bench 2.1: 85.8%, narrowly trailing GPT-5.6 Terra’s 87.4%.
• AutomationBench: 30.4%, crushing Gemini 3.6 Flash’s 17% and Claude Sonnet 5’s 10.7%.

Translation: Google’s lightweight model just walked into the heavyweight division wearing sunglasses and started swinging.

It Isn’t Just a Coding Machine

The model also delivered some huge gains outside software development. Gemini 3.7 Flash scored 34% on GDP.pdf, a difficult test involving expert-level PDF comprehension. Gemini 3.6 Flash managed only 22%, while Claude Sonnet 5 scored 28% and GPT-5.6 Terra reached 24.7%.

It posted a massive 97% on Google’s 128,000-token long-context test and scored 85.4% in long-video understanding. Google says the model can process text, images, audio and video with a context window of up to one million tokens.

The company is pitching 3.7 Flash as a model capable of debugging software, operating tools and completing complicated business workflows with less human supervision. Reuters reports that Google is also offering an introductory API price of 75 cents per million input tokens and $3.75 per million output tokens through the end of 2026.

That combination of intelligence, speed and relatively low pricing could make Gemini 3.7 Flash a major headache for competitors charging considerably more.

So, How Powerful Could Gemini 4 Pro Be?

Here is where things get spicy.

Gemini 3.7 Flash is based on the existing Gemini 3.6 foundation. It is not the result of Google’s enormous new Gemini 4 training run. If algorithmic improvements can push a Flash-class model this far, an entirely new and significantly larger frontier model could deliver a much bigger jump.

Google CEO Sundar Pichai has confirmed that Gemini 4 is currently being trained and said the company is putting substantial computing power and effort behind it. He also indicated that the next frontier generation will require a larger base model, according to comments documented by 9to5Google.

That does not guarantee Gemini 4 will dominate every benchmark. Bigger models can still suffer from slow responses, excessive token usage, hallucinations and inconsistent real-world performance. Benchmark results supplied by model developers should also be treated as evidence—not a final verdict.

Still, the trajectory is hard to ignore. Google appears to have squeezed flagship-level ability into its Flash line. Giving its engineers a larger foundation, more compute and additional training time could create something seriously formidable.

Is Google About to Skip Gemini 3.5 Pro?

Gemini 3.5 Pro has become the mysterious missing celebrity of Google’s AI lineup. The company previously said the model was being tested with partners and would arrive “soon,” but Gemini 3.6 Flash and now Gemini 3.7 Flash have both appeared without a firm 3.5 Pro launch date.

Axios has raised the possibility that Google could skip the delayed release and make Gemini 4 Pro its next major flagship. That remains speculation, however, and Google has not announced that Gemini 3.5 Pro is canceled.

Rumors currently point toward a possible late-2026 Gemini 4 reveal, with November or December being floated based largely on Google’s previous release patterns. There is no confirmed launch date, official benchmark sheet or finalized “Gemini 4.0 Pro” product name yet.

The Bottom Line

Google hasn’t officially unleashed Gemini 4 Pro—but Gemini 3.7 Flash just sent one heck of a warning shot.

A supposedly efficient workhorse model is already beating flagship competitors in several coding, automation and document-understanding tests. If Google can carry that momentum into its larger Gemini 4 foundation, the result could be the beast developers have been waiting for.

OpenAI, Anthropic and the rest of the AI heavyweight club might want to keep one eye on Mountain View—because Google’s next Gemini may not be coming to participate.

It may be coming for the crown.