The AI circle has been as lively as a market these days. It’s just a blast! When everyone was asleep in the middle of the night, OpenAI suddenly released a “king bomb” online, releasing a new GPT-4.1 series of models. Before anyone could even react, another of the AI giants, Zhipu AI immediately announced the launch of IPO guidance, clearly intending to make a big splash in the capital market. The most worrying thing is DeepSeek and its boss Liang Wenfeng. Since the R1 model came out at the beginning of this year, it has ignited the entire AI circle. Everyone is craning their necks, eagerly waiting for them to make another big move!
Competition in the AI circle is so fierce now, just like a boxing match, one after another. The new model launched by OpenAl this time, under the banner of lower cost and faster speed, is clearly intended to compete with DeepSeek; Zhipu AI is also not idle, and has plunged into the listing process, just to make more money to continue research and development. But there is no news about Liang Wenfeng, which makes people wonder: What big move is he planning? Is he going to wait until everyone is about to forget about him before making a killer move?
OpenAI sneak attack: The new model is so crazy, and is it forced to give in?
In the early morning of April 15, when everyone was dreaming, OpenAI suddenly released a blockbuster news online – the launch of the GPT-4.1 series of models! This series includes GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano. Just by listening to the name, you can tell that the main feature is “small but powerful”. The official propaganda said that the biggest selling point of these new models is lower cost and faster speed. It’s simply here to ruin the show!
How powerful is it? Take GPT-4.1 Mini for example. In several professional tests, its performance is better than that of the previous GPT-40, with significantly reduced latency and 83% cost reduction! GPT-4.1 Nano is even more amazing, capable of processing context windows of up to 1 million tokens. Tasks such as classification and auto-completion are a piece of cake for it. Moreover, these three new models have increased the context window processing scale from the original GPT-40’s 128K to 1 million tokens, which has greatly upgraded its ability to understand long texts!
In terms of price, OpenAI has invested heavily this time. GPT-4.1 is 26% cheaper than GPT-40, with input and output costs of only $2 and $8 per million tokens respectively; GPT-4.1 Nano is even more powerful, with input and output costs of only $0.1 and $0.4. Let’s take a look at DeepSeek. The cost of input and output of one million tokens for the V3 model is 0.5 yuan and 8 yuan, and the cost of R1 inference is 1 yuan and 16 yuan respectively. Now everyone can see that OpenAI is determined to fight a price war with DeepSeek, and it’s just a matter of writing “I want to grab the market” on its forehead!
But surprisingly, OpenAI has suspended GPT-4.5, which was just released in February. Why? The reason is simple: it is too expensive! The API pricing of GPT-4.5 is as high as $75 and $150 per million tokens for input and output. Who can afford this price? The official explanation is that it is to concentrate resources on future model development. But everyone knows that DeepSeek’s low-price strategy forced OpenAI to give in and adjust its strategy.
What’s more interesting is that OpenAI’s CEO Sam Altman used to hate DeepSeek, accusing it of using GPT’s data, and wrote to the US government, demanding restrictions on the flow of AI technology to China, and even wanted to ban DeepSeek directly. But now, he has to admit that DeepSeek has indeed affected the development of OpenAI, and he also said that he would launch a better open source model. The once tough OpenAI has now been forced to change its strategy by DeepSeek, which shows how great the influence of DeepSeek is!
Zhipu AI has a big move: work hard and go straight to the market!
On the same day that OpenAI released its new model, Zhipu AI was also not idle and directly announced the launch of its IPO counseling filing! The counseling institution is the well-known CICC, and their plan has been finalized. The IPO counseling plan will be completed from August to October this year. This speed is faster than a rocket!
In fact, Zhipu AI has already revealed its “ambition”. At the end of March, Zhang Peng, CEO of Zhipu AI, said that there is still a long way to go to achieve general artificial intelligence (AGI). The money raised now is just the money for the journey, and more funds will be needed for research and development in the future. Now it seems that going public is their big move to make money. Once the listing is successful, they will get a large amount of funds, and then they will be able to be more generous in AI research and development, and compete with giants such as OpenAI and DeepSeek!
Although Zhipu AI is not so high-profile in front of the public, its strength should not be underestimated. The sudden announcement of its listing is like a bomb dropped in the AI circle, making everyone realize that the competition in the AI field is not just a two-person show between OpenAI and DeepSeek, but also dark horses like Zhipu AI, ready to rush out and share a piece of the pie!
DeepSeek plays mysterious: What is Liang Wenfeng up to?
If we talk about the most confusing things in the AI circle recently, it must be DeepSeek and Liang Wenfeng. Since the release of the R1 model on January 20 this year, DeepSeek has been like a super star, instantly illuminating the entire AI circle. The performance of the R1 model can directly compete with OpenAI’s o1 official version, and it is completely open source. This wave of operations directly put DeepSeek on the altar.
Since then, everyone has been looking forward to DeepSeek’s new moves every day. According to DeepSeek’s previous product update rules, it will be upgraded almost every 3 months. V2.5 will be released in September 2024, V3 will be released in December, and V3 will be updated to V3-0324 in March 2025. Some people predict that DeepSeek’s next update to the R1 model may be in early May, or even earlier. But until now, DeepSeek officials have not revealed a word, and this sense of mystery can drive people crazy
But DeepSeek is not completely silent. In March, the official website suddenly posted information about recruiting CFO, COO, and financial director, which immediately triggered everyone’s speculation: Is DeepSeek also going to learn from Zhipu AI and prepare for financing and listing? Then, on March 25, they released an updated version of the V2 model. Although it was not as sensational as the R1 model, it also showed that the team has been working hard on research and development. On April 3, Liang Wenfeng and the team also published a paper specifically studying how to make the model more efficient in processing long texts. These small actions are telling everyone that DeepSeek is not idle and has been secretly releasing big tricks!
But Liang Wenfeng also faces many difficulties. On the one hand, foreign giants such as OpenAI and Google are desperately catching up, especially the new model released by OpenAI this time, which is obviously aimed at DeepSeek; on the other hand, the US government is still restricting the export of Nvidia H20 chips to China, but this is exactly the key equipment DeepSeek needs most. Liang Wenfeng said before, “Our biggest difficulty is not the lack of money, but the export ban on high-end chips.” But he was not intimidated, and said, “There is nothing mysterious about Nvidia’s GPU. To catch up with them, you have to rebuild the team, spend time and money, and catch up with the next generation of technology. This is the real competitive advantage.” Although this is a domineering statement, everyone is waiting to see whether it can be done!
AI Circle Melee: Who will have the last laugh in this war?
The current AI circle is like a fierce melee. Giants such as OpenAI, DeepSeek, Zhipu AI, and Google are all desperately engaged in R&D and grabbing the market. Some are constantly launching new models to reduce costs and increase speed; some are diving into the capital market to get more funds. This war has entered a white-hot stage!
From the perspective of industry development, there is still a long way to go to achieve general artificial intelligence (AGI), which requires a lot of capital and technology investment. Moreover, open source seems to have become a major trend. Even OpenAI, which has always been high on the list, was forced by DeepSeek to start open source. This shows that in the field of AI, it is definitely not enough to rely on one’s own closed-door development. Only when everyone shares with each other and makes progress together can the development of the entire industry be promoted.
