OpenAI Cuts Off API Supply to China, China’s Large Model Companies Launch ‘Relocation Plan’
Developers of APIs from various countries and regions, including mainland China, have successively expressed on social platforms that they have received a ‘warning letter’ from OpenAI, stating that additional measures will be taken to stop the use of their API in unsupported regions.
The letter stated: ‘Our data shows that your organization has API traffic from a region that OpenAI does not currently support. We will be taking additional measures to block API traffic from regions that are not on our supported countries and territories list starting on July 9.’
For developers, this is undoubtedly a heavy blow. Many developers rely on OpenAI’s API to build and optimize their applications and services. Once the blocking measures are implemented, these developers will have to look for alternative solutions or try to reconfigure their services in supported regions.
Some developers express concerns about this, fearing that the blocking measures will affect their business development and may even lead to stagnation of some innovative projects. At the same time, there are also views that this will prompt developers to pay more attention to developing localized solutions, thereby promoting technological diversity and innovation.
However, today, major Chinese model companies have taken action immediately: Zhipu AI announced in the afternoon that it will provide a ‘special relocation plan’ for OpenAI API users; while SiliconFlow has made a large number of models free; In addition, 01.AI, MiniMax, and Baidu AI Cloud also released important information such as zero-cost migration and API price reduction.
The overall arrangement is as follows:
Zhipu AI is specifically designed for OpenAI API users, providing a special migration plan to help users easily switch to Chinese large models.
Provide for developers: 150 million tokens (50 million GLM-4 + 100 million GLM-4-Air), a series of migration training from OpenAI to GLM
Provide for high-volume customers: Token gifting plan equivalent to the scale of OpenAI usage (no limit), concurrent scale equivalent to OpenAI, matching high-level membership policies, exclusive relocation consultants and support from five technical experts per day, as well as filing training and assistance.
It is reported that the GLM series of large models by Zhipu AI has now fully benchmarked against the OpenAI product system. The GLM-4 model ranks higher than models such as Llama3-70B-Instruct and Claude 3 Sonnet.
SEE ALSO: Zhipu AI Unveils GLM-4 Model with Advanced Performance Paralleling GPT-4
On the 25th, domestic AI infrastructure company SiliconFlow announced its “Token (Character Token) Free” plan: top open-source large models such as Qwen2-7B, GLM-4-9B, Yi-1.5-9B are now permanently free.
The prices of large model APIs on SiliconCloud are also very cheap. Even for Qwen2-72B-Instruct, the official website shows that it costs only 4.13 yuan per 1M Token. New users can also enjoy 20 million Tokens for free.
It is reported that SiliconFlow was founded in August 2023, positioning itself in the AI infrastructure layer of the era of large models. The company aims to accelerate the development of General Artificial Intelligence (AGI) by reducing the cost and development threshold of large model applications through technological innovation. Under this goal, SiliconFlow’s R&D team has the capability to develop distributed deep learning frameworks and continuously introduce innovative AI infrastructure products to meet market demands.
On the afternoon of June 25th, Baidu AI Cloud Qianfan launched the Large Model Inclusive Plan, claiming to switch to the largest domestic model platform with zero cost. Baidu AI Cloud emphasized that the scenarios of Baidu AI Cloud Qianfan are more diverse, models are more comprehensive, toolchains are more complete and easy to use, and it is safer and more reliable.
Baidu AI Cloud announced that starting today, it will provide the following services to newly registered enterprise users:
Zero-cost access: The ERNIE flagship model is free for the first time, with a gift of 50 million Tokens package for the ERNIE3.5 flagship model, and continuous free access to main models ERNIE Speed/ERNIE Lite and lightweight model ERNIE Tiny; additional gifts of ERNIE3.5 flagship model Tokens package equivalent to OpenAI usage scale for users migrating from OpenAI.
Zero-cost training: Free fine-tuning training service for models
Zero-cost migration: Zero-cost SDK migration tool
Zero-cost services: Expert services (migration & usage guidance)
However, Baidu AI Cloud also stated that the above promotional activities are applicable until 24:00 on July 25, 2024.
On the evening of June 25th at 18:00, 01.AI, a unicorn company founded by Dr. Kai-Fu Lee, announced the launch of the ‘Yi API Half-Price Replacement Plan’, targeting OpenAI users to provide services for smooth migration to Yi series large models. For users accessing different models from OpenAI, 01.AI has correspondingly provided high-performance and cost-effective replacement solutions.
New customers currently registered to use the Yi API will immediately receive a credit limit of 100 yuan, helping users make a smooth transition; platform top-ups will also receive a 50% bonus credited amount, with no upper limit, providing users with longer-term benefits; any top-up will enjoy RPM/TPM speed limits directly upgraded to Tier3, reaching advanced service quality and ultra-fast response speeds; in addition, the Yi API will also provide Prompt-compatible tuning services support to help users adapt quickly and well to the Yi series large models.
On the afternoon of June 25th, MiniMax launched a zero-cost nanny-style migration solution. The open platform interface is fully compatible with OpenAl, enabling one-click migration and free use of the MOE architecture large model abab 6.5s per month without restricting TPM RPM. Technical expert support is provided during the migration process, with responses to questions available at any time.
MoonShot AI previously revealed that in order to facilitate AI developers to quickly access model capabilities, the API of Moonshot AI’s open platform is compatible with OpenAI. Developers can smoothly migrate without making any “extra” modifications to the code other than basic parameters, and immediately experience the capabilities of Moonshot models.
However, in reality, the shutdown of the OpenAI API has limited impact on China. The main reason is that large model websites in China need to go through record-filing and approval processes. Many Chinese companies relying on OpenAI technology deploy GPT APIs on overseas sites. Therefore, for China, what should be more vigilant about are the risks of using models brought by the closure of American open-source large models such as Llama, Phi-3, BLOOM, and Gemma which may widen the technological gap between the two countries.
China Fund News commented that there are currently two main channels for using OpenAI technology in China, and the API capabilities provided by them are no different: First, connect to the API provided by OpenAI official. (Currently, it may be banned at any time when used domestically; individual interest developers can choose); Second, connect to Microsoft Azure’s OpenAI technology. (Currently the only compliant channel in China, application requires enterprise qualifications).
From a different perspective, the cessation of OpenAI’s API services to China may have a certain short-term impact on China’s AI industry. However, in the long run, it could also become an opportunity to promote domestic AI technology independent innovation and research and development. It may prompt Chinese large-scale model companies to accelerate independent research and development, promote the development and innovation of domestic AI technology, and lead more developers and enterprises domestically towards using domestic large models to avoid potential risks in the future.
