The catfish effect is beginning to show: OpenAI launches the o3 model to counter DeepSeek

Just as DeepSeek's momentum continues to soar, OpenAI has been gearing up, and its countermeasures have finally arrived.

 

A few days ago, OpenAI urgently launched the OpenAI o3-mini series of models, claiming to be the latest and most cost-effective model in the current inference series.

In this release, the o3-mini comes in three versions: low, medium, and high. Its selling points are higher output quality than o1, but the price is only one-tenth of o1, and the context window has been increased to 200,000 tokens.

 

According to the official statement, the goal of the o3 series of models is to push the boundaries of low-cost inference.

 

This is also the first inference model that OpenAI has launched for free to users. Even in the Reddit "Ask Me Anything" event, CEO Altman publicly reflected, which is rare: "Personally, I think we were on the wrong side of history when it comes to the issue of open-sourcing AI model weights."

 

So,let's take a closer look.

 

Let's first look at a graph.

This is a relatively authoritative exam in the AI field-"The Last Human Exam".

 

 

At the same time, o3-mini achieved the highest scores in benchmark tests such as mathematics and code.

 

Especially in the high-difficulty "physical simulation" challenge, o3-mini directly defeated R1, demonstrating its powerful capabilities.

 

It can be seen that o3-mini has more advantages in physical reasoning.

 

From some online tests by netizens, o3-mini also performed perfectly when dealing with doctoral-level scientific problems.

 

However, DeepSeek did not perform as well when dealing with some problems in the fields of physics and mathematics.

 

Although the new o3-mini model of OpenAI has improved in performance, it has not brought a revolutionary change in cost like DeepSeek-V3.

 

At the API level, for o3-mini, the input cost is $1.10 per million tokens and the output cost is $4.40 per million tokens. Although the price is 63% cheaper than o1-mini and 93% cheaper than the full-featured o1, it is still about 7 times that of GPT-4 0mini.

 

This means that for large-scale applications and training, the cost of OpenAI's new model is relatively high.

 

For enterprises and developers, cost is an important consideration when choosing a model. In this regard, DeepSeek-V3 is obviously more attractive.

 

 

As I said before, the emergence of DeepSeek is more like a catfish, prompting existing AI giants to work harder.

 

The competition between OpenAI and DeepSeek is just a microcosm of the fierce competition in the AI industry.

 

With the continuous development of technology, more and more enterprises and institutions are joining the AI field.

 

In the future, the AI industry may show a trend of flourishing development.

 

One hand, emerging forces like DeepSeek continuously challenge industry giants through technological innovation, promoting the technological progress and cost reduction of the entire industry.

 

 

 

At the same time, in the future, AI technology may be applied in more fields. From content creation, intelligent customer service to industries such as healthcare and finance, all will be deeply influenced by AI.

 

AI will also become an important force in promoting the digital transformation of various industries.

 

Regarding the release of o3-mini, I'm glad to see such competition.I'm looking forward to the day when AI completely changes the world.