Google Gemini 2.0 Pro is here, is DeepSeek suddenly not as popular anymore?

Google's major update has made the AI ​​circle excited again! Gemini 2.0 Pro Experimental was officially released. This is Google's most powerful Gemini model to date.

At the same time, Google also brought two new members: Gemini 2.0 Flash and Gemini 2.0 Flash-Lite Preview. It seems that Google is planning to make a large-scale "sweep" in the AI ​​battlefield!

Model parameter comparison

Let's first talk about the performance of DeepSeek V3. Its context length is 64K and its maximum output is 8K, which is not the best, but it is enough.

The input cost was $0.27/million tokens, and the output cost was $1.1/million tokens, which seemed very competitive at the time. However, it only supports text processing, which seems a bit thin in this era of multimodality.

Gemini 2.0 Flash brought astonishing data: the context length and maximum output both reached an astonishing 1M, several times higher than DeepSeek V3. What's more interesting is that its input cost is only $0.1/million tokens, and its output cost is $0.4/million tokens, which directly beats DeepSeek V3.

The most surprising thing is that Gemini 2.0 Flash not only supports text, but also supports image, audio and video processing .

This is like using a plug-in. A model can handle so many types of data. It is truly an "all-rounder"!

Although its first token response delay reaches 1.32 seconds, which is slightly slower than DeepSeek V3's 1.17 seconds, its throughput is as high as 159.7 tokens/second , completely crushing DeepSeek V3's 8.95 tokens/second.

free!

During the Gemini 2.0 Flash (experimental preview) period, you can use it for free. However, please note that there may be some limitations to free use, such as a limit of 1,000 requests per day when using the Search tool .

How to use Gemini 2.0 Flash?

Gemini 2.0 Flash can be used in two main ways:

Google AI Studio (web interface): This is a user-friendly way to experience Gemini 2.0 Flash without programming experience.

Address: https://aistudio.google.com/

Gemini API: This is more for developers, allowing you to integrate Gemini 2.0 Flash into your own applications.

My personal test report is here: Too elegant! Gemini2.0 SDK first test: It can read and read, and it has built-in search function. The key is that it is free!

Google announced that Gemini 2.0 Flash is free to use during the experimental preview period! Although there is a limit of 1,000 requests per day, it is enough for most users. Whether through the web interface of Google AI Studio or through the API to access your own application, you can easily get started.

Paid version of Gemini

However, I personally subscribed to the paid version of Gemini, which costs $20 per month. When I opened the page today, I saw all the major updates:

The more useful one is 2.0 Flash Thinking Experimental with apps, which can directly penetrate the entire Google ecosystem and completely catch everything. For actual experience evaluation reports, please pay attention to the latest articles on the Byte Notebook WeChat public account

Here is a small secret: other DeepReseach do not have the function of self-checking answers, that is, version 2.0 can also correct the content after deep search. For more advanced usage skills, please pay attention to the latest articles!

DeepSeek is actually not bad

Of course, DeepSeek also has its own advantages. For example, it may perform better in text processing in certain professional fields. But in terms of overall strength, especially in terms of cost-effectiveness and functional diversity, the emergence of Gemini 2.0 has indeed brought a considerable impact to the market.

at last

It seems that the competition in the field of AI can never stop! One second you are still marveling at the breakthrough of a certain model, and the next second a new player enters with stronger strength. For users, this is definitely a good thing, after all, competition can bring progress and allow us to use better AI services.