Google has finally updated its model; the combination of software and hardware is its greatest advantage.

Yesterday in the early hours, Google finally released its latest large model, Gemini 3.5. It also released a series of the latest data for its large model, which is also very impressive:

Gemini App's monthly active users have doubled in one year to over 900 million, Google Search's AI Overview feature users have exceeded 2.5 billion, and AI mode users have surpassed 1 billion.

This release of 3.5 is a significant update following the 3.1 release half a year ago. Gemini 3.5 Flash focuses on speed and low cost and will be the default model for Gemini App and Search AI mode. Additionally, 3.5 Pro will be officially released in June.

The improvements in specific parameters and performance are divided into two levels.

One is the improvement compared to its own pre-3.1 versions, and the other level is the comparison among large models in the entire market. Overall, Google's large model is considered above average, but definitely not the best. Combined with my personal usage experience, there is still a gap compared to Claude and GPT. However, Google's large model excels in its excellent integration with its vast Google ecosystem, such as with YouTube. In fact, except for a small number of people doing programming and high-end scientific research, for most ordinary users, being able to utilize and make good use of Gemini's model capabilities is sufficient.

In addition to these large models and the soft aspects of its own ecosystem, Google has also taken a differentiated path in the hardware aspect, specifically in the development of TPU chips compared to GPUs.

Therefore, this path taken by Google is something that domestic companies like Alibaba and Tencent can refer to and learn from. There's no need to make the model the absolute best, but through self-developed chips + above-average model capabilities + relying on a vast own internet ecosystem, in the new AI era, they should still be able to consolidate their own moat position.

(Not as investment advice)

$Alphabet(GOOGL.US)  $BABA-WR(89988.HK)  $TENCENT(00700.HK)  $Penghua Guozheng Semiconductor Chip ETF(159813.SZ)