The "landing race" has begun.

Recently, many developers using OpenAI have received official emails stating that they have detected API traffic from unsupported regions in their accounts. After politely explaining the official policy, OpenAI bluntly stated:

From July 9th, we will take additional measures to block API traffic from regions not included in the list of supported countries and regions. To continue using OpenAI's services, you need to access the service from a supported region.

Source: OpenAI

Currently, China is not included in the 188 countries and regions supported by OpenAI, which means developers in these regions will need to take on more additional risks if they want to continue accessing OpenAI's API. According to OpenAI's terms of use, accounts that violate the terms will face various restrictions, such as service suspension and account bans.

OpenAI's ban on violating accounts has not started recently. In fact, shortly after the release of ChatGPT 4.0, OpenAI had already intermittently banned some accounts from restricted regions, but for API traffic accessed through server jumps or other methods, OpenAI adopted a turn-a-blind-eye attitude.

What was once an unspoken understanding has suddenly ended, leaving many developers caught off guard. Some developers have even直言, OpenAI's warning letter has put their projects at risk of suspension, and they will consider practical measures based on specific circumstances.

How will OpenAI's move affect AI development in China?

More worried than developers are the "shell large models"

Although many domestic AI models claim to have reached or approached the level of ChatGPT 4.0, it must be acknowledged that ChatGPT 4.0 is still the best AI large model globally, with no real competitors in the general domain for the time being.

However, in the Chinese domain and some specific domains, domestic AI models like Wenxin Yiyan and Tongyi Qianwen do possess the strength to compete with ChatGPT 4.0. For most domestic AI application developers, these AI large models can meet their daily development needs.

However, the companies with the greatest demand for OpenAI are actually AI large model development companies. Currently, in addition to leading companies like Baidu that are promoting the research and development of AI large models, a large number of start-ups are also entering the AI large model race. To efficiently train their AI large models, they often choose to use ChatGPT 4.0's answers as reference and adjust their AI large models accordingly.

Source: OpenAI

In these companies' projects, ChatGPT 4.0 actually exists more as a "mentor." OpenAI's ban measures will undoubtedly impact these companies' AI large model training projects, resulting in project delays and the need to invest more server resources and manpower.

Moreover, the most severely impacted may be some "shell" AI large models. As one of the hottest areas in the investment market, not all AI large models are "genuinely self-developed," and some shell enterprises abound. What is a "shell"? Simply put, it is packaging ChatGPT's answers as one's own AI large model answers through some means to deceive investors.

These companies tend to have the highest dependency on OpenAI because switching to other domestic AI large models may face legal risks. Overseas AI large models like Gemini and Claude do not perform significantly better than domestic AI large models and are far less attractive to investors than "ChatGPT." In this sense, OpenAI may help clear out some substandard AI enterprises in China.

OpenAI's ban will indeed cause some trouble for domestic AI developers. However, OpenAI is not irreplaceable. In common domains, domestic AI large models can now achieve 80% to 90% substitution. With subsequent iterations and upgrades of domestic AI large models, we believe the effect will become even better.

Although many people are angry about OpenAI's actions, there's no point in trying to please someone who doesn't want to be pleased. Moreover, past experience tells us that self-reliance is the best choice. Instead of trying to bypass restrictions, it is better to work with domestic enterprises to make domestic AI bigger and stronger.

So, the biggest question now is: What domestic AI large models can replace OpenAI? What are their respective advantages, costs, and current migration solutions?

"Welcoming Back" Developers, Domestic Large Models Compete

OpenAI's refusal of Chinese developers is a rare opportunity for domestic large model suppliers.

Given that OpenAI is a pioneer, many developers firmly believe that GPT is much more powerful than domestic large models, and domestic large models have not had the opportunity to prove themselves despite their efforts. Additionally, some developers have told Leitech that although they have seen the progress of Zhipu AI, Baichuan, and Kimi, they have not had the motivation to change their underlying large models due to high migration costs.

Now, Chinese developers are forced to migrate back to domestic large models. Seizing this opportunity, many mainstream domestic AI large models have issued "recruitment orders" and offered various preferential policies and migration solutions to enable developers to easily transfer their AI projects "back home." Domestic large models providing relocation services to "welcome back" developers is both a timely assistance and a natural trend. So, which large model will be able to grab the developers abandoned by OpenAI?

1. Wenxin Large Model: Zero-cost migration, free lightweight products

Baidu Smart Cloud has launched a domestic large model inclusive plan called "Hometown Cloud." Starting today (June 25th), it offers the following preferential policies to all newly registered enterprise users: The first call to the Wenxin flagship model is free, and it comes with a 50 million Tokens package for the ERNIE3.5 flagship model. Additionally, it provides free model fine-tuning training, zero-cost SDK migration tools, and expert services.

Source: Leitech

ERNIE3.5 flagship model is an important version of Wenxin's large model, supporting 5K tokens input and 2K tokens output. Compared to the previous version, it has undergone comprehensive optimization and upgrades, meeting the usage needs of complex task scenarios and performing far better than other versions in semantic and contextual understanding, making it more suitable for scenarios like virtual assistants, online customer service, and social media.

In comprehensive ability tests, the ERNIE3.5 flagship model scored higher than ChatGPT 3.5 and surpassed ChatGPT 4.0 in Chinese tests. Additionally, thanks to Baidu's authoritative position in the search field, the ERNIE3.5 flagship model outperforms other large models in AI search and other fields, making it highly suitable for AI applications and development projects that require real-time data updates.

Furthermore, Wenxin's large model also provides three completely free AI models: ERNIE Speed, ERNIE Lite, and ERNIE Tiny. These AI models can be deployed in lightweight usage scenarios, saving costs for enterprises and providing conversational and question-answering experiences that exceed traditional intelligent services.

If you want to learn more about Wenxin's large model's relevant solutions, you can visit Baidu Smart Cloud's WeChat official account to view related articles.

2. Tongyi Qianwen: Price is only 1/50 of GPT-4

Alibaba Cloud's OpenAI migration campaign is concise and to the point, with less than 50 words highlighting two key points: 1) The AI large model's strength is not inferior to GPT-4; 2) The price is only 1/50 of GPT-4.

Source: Leitech

Alibaba Cloud also provides exclusive migration services for OpenAI users to ensure developers can efficiently and easily migrate API interfaces to Alibaba Cloud, and it will also provide a free 22 million Tokens quota. As an enhanced version of Tongyi's ultra-large language model, Qwen-plus's API access supports up to 30K Tokens of text input, better handling long-form content.

At the same time, Alibaba Cloud's Bailian large model platform also provides different AI models for various application scenarios, supporting multiple uses from text-to-image, AI customer service, AI office, to AI video generation. Using dedicated AI large models can not only improve AI creation quality but also reduce corresponding costs.

In addition, Alibaba Cloud also provides services such as AI large model hosting. Interested friends can visit Alibaba Cloud's WeChat official account or official website for more information.

3. iFLYTEK Spark: Free Lite version, free Tokens

iFLYTEK Spark is also a well-known domestic AI large model, ranking among the top in AI voice, translation, and other fields. This time, coinciding with the official release of iFLYTEK Spark V4.0, iFLYTEK Spark has launched a new inclusive policy: Three different models, Spark Lite, Pro, and Max, offering permanent free, one-month free, and 100 million Tokens giveaway activities, respectively. Even the latest Spark 4.0 Ultra version also offers a buy-one-get-one promotion.

Source: Leitech

It must be said that iFLYTEK Spark's sincerity is almost overflowing the screen. In addition to the latest Spark 4.0 Ultra (comparable to GPT-4 Turbo), all other Spark large models provide a significant amount of free quota, with Spark Lite announcing permanent free usage, allowing developers to deploy lightweight AI applications at zero cost.

Spark Max's comprehensive ability in the Chinese domain is not inferior to GPT-4, and its adjusted price is only 0.21 yuan per 10,000 Tokens, allowing you to generate text content equivalent to Yu Hua's "To Live" for only 2.1 yuan.

Additionally, the iFLYTEK Spark V4.0 released on June 27th will bring developers and users an even better experience and more powerful AI large models, while the price will still be significantly lower than OpenAI. If you have complex task needs, you can keep an eye on the launch and subsequent pricing plans.

4. Zhipu Qingyan: I can offer what OpenAI can

Although Zhipu Qingyan is not as well-known as Wenxin or Tongyi in application fields, it is a regular on various AI large model test rankings. This time, it has launched an exclusive relocation plan for OpenAI developers.

Zhipu Qingyan provides each developer with a free usage quota of 150 million Tokens (50 million GLM-4 + 100 million GLM-4-Air) and corresponding migration training to ensure developers can easily migrate their applications to the GLM large model platform.

Source: Leitech

For users with higher demand, Zhipu Qingyan also provides additional benefits, such as a Token giveaway plan equivalent to OpenAI's usage scale (no upper limit) and concurrent scale equivalent to OpenAI, as well as dedicated technical consultants and expert support.

As a leading domestic AI large model, the GLM-4 large model supports up to 128K context window length, easily handling long text questions and answers, and supports multi-modal data processing such as images and sounds, making it perform very well as a general-purpose large model.

Interested friends can visit Zhipu's WeChat official account to view related information and add customer service for consultation.

5. Baichuan Large Model: API compatible with OpenAI

Recently, Baichuan Large Model released a new generation of Baichuan 4, with performance ranking among the top in China, rivaling overseas mainstream AI large models. It provides diverse AI model options, including Baichuan 4 and opening up multiple API interfaces such as Baichuan3-Turbo, Baichuan3-Turbo-128k, and Assistant API.

Source: Leitech

To facilitate the migration of OpenAI developers, in addition to providing a free quota of 10 million Tokens, Baichuan Large Model also offers free usage rights for Assistant API and launches corresponding technical expert Q&A services, enabling API migration to be completed in as little as 5 minutes.

Based on the information obtained, Baichuan 4 has improved its general capabilities by 10% compared to Baichuan 3, with 14% and 9% improvements in digital and coding abilities, respectively. It also possesses excellent multi-modal processing capabilities, enabling it to provide better services for multi-purpose AI applications and development projects.

Moreover, Baichuan 4 has prepared a more user-friendly development platform for developers and will launch a zero-code Agent creation platform product in the future, allowing developers to create Agents through natural language, even for users who do not understand code, further lowering the entry barrier in the AI field.

It is worth noting that Baichuan Large Model's preferential activities are targeted at all developers, whether you are an OpenAI migration user or a new developer, you can visit Baichuan Large Model's WeChat official account for details.

6. Tiangong Large Model: Stronger Chinese capabilities than GPT-4

For migrating OpenAI developers, Tiangong Large Model also provides a series of preferential policies: free 150 million Tokens and exclusive migration service technical support, filing training, and assistance, providing developers with more comprehensive services.

Source: Leitech

Tiangong AI Large Model is a self-developed ultra-large language model by Kunlun Wanwei. In relevant tests, Tiangong 3.0 achieved excellent scores, and its actual performance in Chinese tasks even surpassed GPT-4. Moreover, Tiangong 3.0 is currently one of the world's largest and most powerful open-source MoE models in terms of model parameters and performance.

For developers with complex task needs, Tiangong 3.0's multi-modal processing capabilities and long text processing capabilities are sufficient to handle different scenario requirements. For developers with more lightweight usage scenarios, they can also choose the cost-effective Tiangong 2.0 large model, whose actual performance is not inferior to ChatGPT 3.5 and also possesses industry-leading AI performance.

Interested friends can visit the Tiangong AI Smart Assistant WeChat official account for consultation.

7. Tencent Hunyuan Large Model: New users get 100 million Tokens

As one of the BAT giants, Tencent is obviously not absent. According to the latest news, Tencent Cloud is offering 100 million Tokens for Tencent Hunyuan Large Model to OpenAI and newly migrated enterprise users, along with free dedicated migration tools and services.

Source: Leitech

In various tests at home and abroad, Tencent's Hunyuan Large Model has achieved outstanding results. Coupled with Tencent's years of technical accumulation in the AI field, it can provide developers with dedicated AI large models of various scales and different domains, further reducing developers' usage costs and development difficulties.

In addition, Tencent is expected to launch a series of AI game development tools in the future. For game developers and related practitioners, Tencent's Hunyuan Large Model's performance in the entertainment field is even more值得期待. Interested friends can visit Tencent Cloud's WeChat official account for more information.

The large model war has fully entered the second half.

Who would have thought that OpenAI's ban would spark a thousand waves, causing a host of domestic AI large models to