Google is making progress on all four fronts: One line of SQL handles data analysis, and medical AI appears in "Nature"

While other manufacturers were still building momentum for a single model, Google chose to light up four fronts at the same time in one day. On July 28, Google, the company behind Gemini, released four products in one go - from data analysis to medical diagnosis, from coding speed up to image generation. The most disruptive one is TabFM, which claims to be "one line of SQL for data analysis".

TabFM: Leveling the threshold of data analysis

TabFM's official positioning is the world's first zero-sample basic model for tabular data. Zero samples mean that users do not need training data or fine-tuning. As long as they can write a sentence of SQL, they can complete table analysis tasks that used to require the intervention of data analysts. More importantly, it has been integrated into BigQuery, Google's enterprise-level data warehouse. This is equivalent to welding the "basic model + enterprise data infrastructure" directly, allowing non-technical users to fish in the deep waters of enterprise data.

The strategic intention of this step is obvious: data analysis is one of the most valuable entrances to enterprise data assets. Whoever can lower the entry threshold will have the say in the implementation of enterprise AI.

Academic Endorsement and Product Matrix

The other line is the academic route. The relevant research results of AMIE, the medical conversational AI system, have been published in the official issue of "Nature" - this is not only a milestone in the direction of Google's medical large model, but also provides endorsement at the top academic journal level for "AI entering clinical practice". At the same time, the coding scene welcomes Gemini 3.5 Flash, which continues the high-throughput and low-latency positioning of the Flash series and targets the developer community; on the image side, there is Nano Banana 2 Lite, which focuses on 4-second generation, serving response-sensitive generation scenarios with light weight and speed.

From an industry perspective, the essence of Google's release is a competition of "ecological breadth": model capabilities, vertical industry applications, and developer tools are advanced in parallel, and the density of the product matrix is ​​used to offset the lead in a single technology point. Compared with OpenAI behind ChatGPT and Anthropic behind Claude which focus on general intelligence and scientific research, Google is betting on the "ubiquitous infrastructure" route and is using specific entrances such as BigQuery, medical care, and development tools to accumulate irreplaceability. This also reminds domestic manufacturers that the winner of AI competition is shifting from "whose model parameters are more" to "whose product reaches a wider range of scenarios."

Several directions worth tracking in the future:

1. TabFM’s actual accuracy in BigQuery: The real boundary of zero-sample capabilities determines whether enterprises can trust it to hand over analysis tasks.
2. AMIE’s distance from paper to clinical practice: After the endorsement by Nature, regulatory approval and hospital implementation are the real thresholds.
3. Gemini 3.5 Flash coding test: In addition to speed, code quality is the winner of developer retention.
4. Positioning of Nano Banana 2 Lite: Can the 4-second image output generate scenes in real time and open up new application space.