Ranking of Domestic AI Large Model Companies
-
When future generations look back on history, they may refer to 2023 as the Year of AI. AI large models have burst into our lives in a disruptive manner, advancing at an astonishing pace—from Q&A dialogues to programming assistance, from image analysis to autonomous creation. The capabilities demonstrated by AI have exceeded most people's expectations, leading many to exclaim, 'The future belongs to AI.'
With the release of ChatGPT, major global tech companies have accelerated the development of AI large models, bringing infinite possibilities to this emerging field. Next, we will highlight ten leading AI large models from the past year, ranked for their innovation, application domains, and AI performance, allowing everyone to easily understand the cutting-edge capabilities of current AI large models. (The following rankings are in no particular order and are arranged alphabetically by pinyin).
ChatGPT is hailed as the progenitor of AI large models. Developed by OpenAI and first released in November 2022, as the first phenomenon-level AI model, ChatGPT has now evolved to version 4.0, boasting processing capabilities far beyond ChatGPT 3.5 and supporting features like plugin loading and real-time internet connectivity.
Currently, thousands of plugins with diverse functionalities have been designed for ChatGPT 4.0, ranging from programming to painting, poetry to music composition—ChatGPT excels at it all. Of course, these capabilities are primarily due to ChatGPT being a multimodal AI model capable of recognizing and analyzing various data types.
ChatGPT remains the most powerful AI model accessible to most people today.
The name Claude might be unfamiliar, but its team has notable origins. Founder Dario Amodei, former VP of OpenAI, led a group of employees to establish Anthropic. After a series of improvements and upgrades, Claude is now regarded as the second-best AI model after ChatGPT, particularly excelling in processing long-form text. After recent upgrades, Claude can now support up to 200k text input, far exceeding ChatGPT 4.0 Turbo's 128K. However, compared to ChatGPT's active user ecosystem, Claude has relatively fewer users and limited application scope.
During OpenAI's internal turmoil, Claude became a backup choice for many enterprises. Since most of Claude's core team members come from OpenAI, it can be considered a weaker version of ChatGPT in some ways. While not as powerful overall, Claude has certain advantages in long-text reading and answering.
From a release timeline perspective, Gemini is undoubtedly the "newcomer" in the AI model field. But according to official descriptions, its capabilities might only be rivaled by ChatGPT. Gemini is Google AI Lab's creation and a multimodal model capable of processing text, audio, images, and other data simultaneously, giving it excellent multitasking abilities.
However, Gemini quickly became a laughingstock after launch when it claimed to be Baidu's ERNIE Bot (in Chinese contexts) or OpenAI's ChatGPT (in English contexts). Although Google subsequently performed emergency maintenance to patch the issue, it still sparked heated discussions online.
Currently in testing phase, whether Google's multimodal large model can surpass ChatGPT remains unknown. But from user experience, Gemini combines advantages of many Google AI models and might bring pleasant surprises in practical use.
While Huawei Cloud's Pangu AI model isn't well-known among general public, it has achieved the farthest and widest commercialization in industries. The Pangu AI model collection includes NLP models, CV models, scientific computing models, etc., making it applicable to scientific research, education, transportation, industry, and other sectors. To maximize the performance of AI large models, Huawei leverages its industrial and product strengths to deploy the Pangu AI model in sectors such as mining, government affairs, finance, and meteorology, with multiple applications already in place. Additionally, Pangu is the first Chinese large-scale model with over a trillion training parameters, ranking among the top in overall capability and leading in industrial applications.
LLaMA 2, developed by Meta, is an open-source AI model offering parameter sizes ranging from 7B to 65B to cater to various user needs. Thanks to its exceptional efficiency, LLaMA 2 delivers solid performance even with lower parameter counts, making it particularly advantageous for localized deployment on small and mobile devices.
Compared to its predecessor, LLaMA 2 shows improvements in multilingual processing, natural conversation capabilities, and long-text comprehension, generating more coherent content with fewer errors. Its open-source nature facilitates rapid adoption and widespread use.
In multiple benchmarks, LLaMA 2 ranks among the top open-source models, making it one of the most popular choices. While it still trails leading models like ChatGPT, it remains an excellent option for AI enthusiasts and small-to-medium enterprises.
Tongyi, a series of Chinese AI large models launched by Alibaba, includes models like Tongyi Qianwen, Tongyi Wanxiang, and Tongyi Lingma. Tongyi Qianwen, for instance, is a natural conversation AI model offering versatile text-based functionalities to address various needs. Tongyi Qianwen, as Alibaba's Chinese AI large model, has advantages in querying business information and other data, with higher accuracy. Currently, Tongyi 2.0 has been upgraded to an AI model with hundreds of billions of parameters, performing better in language understanding and problem judgment. Next, Alibaba plans to integrate it into apps like DingTalk to provide users with real-time online AI support, even supporting diverse functions like AI ordering.
Based on Alibaba's product ecosystem matrix, the Tongyi large model has already been implemented in multiple apps and serves users. It can be said to be the domestic large model most closely related to users. From DingTalk to Taobao, you can see the presence of the Tongyi large model. Interested friends might as well try Taobao's "Taobao Ask" feature, which offers many surprises.
Wenxin Yiyan is a new generation of knowledge-enhanced large language model launched by Baidu. Benefiting from Baidu's long-term investment in AI technology and its leading position in the search engine market, Wenxin Yiyan is the fastest-growing AI large model in China. It excels in various writing, copywriting creativity, Q&A, consulting, coding, and more, providing content that better fits the Chinese context.
Currently, the Wenxin Yiyan web version's "Yiyan Treasure Box" offers rich application scenarios and prompt templates, basically covering the daily needs of domestic netizens. There is also a rich plugin mall, including PPT assistants, long document summary Q&A, academic retrieval assistants, mind maps, and more to expand the capabilities of the large model.
It must be said that Wenxin Yiyan does a great job in understanding user needs, providing many interesting features. Currently, Wenxin Yiyan is free and open, based on the Wenxin 3.5 version. If you want to experience the more powerful Wenxin large model 4.0 version, you need to subscribe to a membership. However, based on personal experience, the cost-performance ratio is still very high.
Xunfei Xinghuo, as the name suggests, is the product of iFlytek. As a leading domestic artificial intelligence enterprise, iFlytek has significant advantages in speech synthesis and recognition, multilingual capabilities, cognitive intelligence, and more. After entering the AI model field, it has fully leveraged its strengths, performing very well in voice conversations, translation, copywriting, and more. Based on the Xinghuo large model, iFlytek is also integrating its various hardware products, allowing users to carry and use the Xinghuo large model on the go. Currently, iFlytek Spark has surpassed the hundred-billion parameter threshold and outperforms ChatGPT in tests, though it still lags behind ChatGPT 4.0. According to iFlytek's plan, the Spark AI model is expected to reach the level of ChatGPT 4.0 by the first half of 2024.
Additionally, iFlytek Spark has launched clients and apps like Content Master to provide users with more direct and convenient solutions, assisting in content publishing, creation, and review. With AI's help, work efficiency is significantly improving.
Youdao Ziyue is an educational large model developed by Youdao specifically for educational scenarios, making it one of the first models to receive national certification. NetEase Youdao has introduced six innovative applications, including large-model translation, essay guidance, grammar tutorials, virtual oral coaches, and AIBox, which have been integrated into various software and hardware products like Youdao Dictionary, Youdao Dictionary Pen, and Youdao Learning Devices.
Ziyue is a rare vertical-specific large model that excels particularly in educational applications compared to general-purpose models. Combining advanced natural language processing technologies, it is especially suited for educational environments, offering personalized learning experiences and enhanced teaching effectiveness.
By integrating the educational large model with software and hardware products, Ziyue's application scenarios have expanded widely, with user experiences surpassing those of general-purpose models in many aspects. Compared to general models, Ziyue provides learners with personalized learning experiences, leading to more efficient learning outcomes. With the growing emphasis on online education, Ziyue's development prospects are very promising.
Although Zhipu AI is less renowned than models from Alibaba, Baidu, or iFlytek, its team is one of the earliest entrants in the AI model field. Zhipu AI released its self-developed GLM-10B billion-parameter model as early as 2020, developed the bilingual trillion-scale pre-training model GLM-130B in 2023, and launched the third-generation base model ChatGLM3 in October this year. It pioneered code interaction capabilities in China, supporting image processing, mathematical calculations, data analysis, and adaptation to over 10 types of domestic hardware like Ascend and Hygon DCU. As one of the earliest open-source AI models in China, Zhipu AI has made significant contributions to the development of domestic AI models and remains a leader in the field. Compared to other AI models, Zhipu AI excels in efficiency, domestic adaptation, and open-source ecosystems, making it a preferred partner for many enterprises.
Over the past year, the development and application of large AI models have reached new heights. These models have not only achieved remarkable technological advancements but also demonstrated broad potential across various fields.
From the versatility of ChatGPT to Claude's long-text processing, Falcon's open-source features, and Huawei's Pangu's commercial applications, each model has contributed uniquely to the advancement and popularization of AI. Whether it's Google's Gemini, Alibaba's Tongyi, Baidu's ERNIE Bot, iFlytek's Spark, or Zhipu AI, these large AI models are continuously pushing the boundaries of technology, enhancing the efficiency and accessibility of artificial intelligence.
The development of AI in 2023 marks the beginning of a new era. These large AI models not only represent the cutting edge of technology but also indicate the direction of future development. As these models continue to evolve and expand their applications, we can expect AI to play a pivotal role in more fields, bringing us more innovation and convenience.