Domestic AI Models

1. Tongyi Qianwen (Alibaba)

• Core Competence: Leading Chinese understanding, outstanding logical reasoning and text creation, supports million-level context windows and multimodal interaction.
• Application Scenarios: Enterprise services, e-commerce, financial customer service, with over 1.5 billion daily calls serving more than 90,000 enterprises.
• Version Status: Multiple iterations, such as Tongyi Qianwen 2.0, continuously optimizing performance, functionality, and multimodal capabilities.

2. Doubao Model (ByteDance)

• Technical Highlights: Nearly 60 million monthly active users, second globally, excels in image understanding and multimodal integration, significant potential in education.
• Cooperation Ecosystem: Collaborates with over 500 enterprises, focusing on family companionship and learning assistance scenarios.
• Version Status: Continuously releases different versions, upgrading in image understanding and multimodal integration to better meet various scenario needs.

3. Wenxin Yiyan 4.0 (Baidu)

• Commercial Advantages: Annual call volume increases by 30 times, with 1.5 billion daily calls, leading in mathematical science and language capability assessments.
• Industry Coverage: Deeply integrates Baidu’s knowledge graph, supporting fields such as healthcare, education, and finance.
• Version Status: Currently focused on version 4.0, with previous versions like Wenxin Yiyan 3.0, each improving knowledge coverage and reasoning capabilities.

4. iFlytek Spark (iFlytek)

• Multilingual Breakthroughs: Supports interaction in over 30 languages, with over 200 million app downloads, mature solutions in healthcare and finance.
• Technical Features: Industry benchmark in speech recognition and synthesis, widely applied in education.
• Version Status: Includes versions 2.0, 3.0, etc., continuously enhancing multilingual interaction and speech capabilities across various industries.

5. Kimi Smart Assistant (Moonlight Dark Side)

• Long Text Processing: Supports input of 200,000 Chinese characters, high popularity in the A-share market, suitable for data analysis and professional document interpretation.
• Scenario Expansion: Plans to extend into legal and research fields.
• Version Status: Continuously updates versions to enhance long text processing capabilities and expand application scenarios.

6. DeepSeek (Deep Search)

• Benchmark in Programming: A complete open-source model ecosystem, R1 version supports code generation and debugging, with comprehensive capabilities comparable to GPT-4.
• Technical Innovations: Breakthroughs in dynamic reasoning optimization and domain adaptation technologies, representing domestic large models on the international stage.
• Version Status: Currently has R1 version, with more optimized versions for code generation and reasoning expected.

7. Zhipu Qingyan GLM-4 (Tsinghua University)

• Interactive Innovations: The first domestic model with video call support, enhancing the naturalness of human-computer interaction.
• Academic Background: Developed by Tsinghua team, balanced capabilities in knowledge Q&A and creative writing.
• Version Status: Developed from the GLM series to GLM-4, with significant improvements in parameter scale and interaction capabilities.

8. Hunyuan Model (Tencent)

• Video Generation: Trillion-parameter scale, supports text-to-video generation, widely used in film and television creation.
• Ecosystem Integration: Deeply integrated into the WeChat ecosystem, providing personalized intelligent agent services.
• Version Status: Continuously updates versions to improve video generation quality and service capabilities within the WeChat ecosystem.

9. Baichuan Model (Baichuan Intelligence)

• Specialization in Healthcare: Solves grassroots medical challenges as an AI doctor, with a disease diagnosis assistance system covering over 1,000 hospitals.
• Open Source Layout: Baichuan-7B/13B models have surpassed one million downloads, performing excellently on evaluation lists.
• Version Status: Includes Baichuan-7B, Baichuan-13B, and other versions with different parameter scales to meet various application needs.

10. Jidream AI (ByteDance)

• Video Creation Tool: Supports generating 1080P videos from text/images, leading in ease of use, deeply integrated into the Douyin ecosystem.
• User Growth: Rapidly popular after its launch in 2024, with a 40% usage rate among short video creators.
• Version Status: Continuously updates versions to optimize video generation effects and user experience.

2025 International AI Model Rankings

1. GPT-4o (OpenAI)

• Developer: OpenAI
• Features: Parameter scale exceeds 10 trillion, supports multimodal inputs (text/image/audio/video), reasoning capabilities close to human levels, excelling in complex logic and cross-domain knowledge integration.
• Application Scenarios: Scientific analysis, cross-industry decision support, and multimedia content generation.
• Version Status: May have different fine-tuned versions for specific applications in various fields.

2. Gemini 2.0 Ultra (Google DeepMind)

• Developer: Google
• Features: Native multimodal architecture, supports real-time translation in over 100 languages, deeply integrated into the Google ecosystem (search/office suite), with a context window extended to 2 million tokens.
• Application Scenarios: Global enterprise collaboration, real-time translation, multimodal search engine optimization.
• Version Status: Includes Gemini 2.0 Ultra version, possibly with lightweight or specific function-optimized versions.

3. Claude 3.5 – Sonnet (Anthropic)

• Developer: Anthropic (Google investment)
• Features: 200K ~ 1M tokens context window, constitutional AI architecture ensures compliance, outstanding performance in healthcare and legal fields, commercialized on-demand billing.
• Application Scenarios: Legal document analysis, medical diagnosis assistance, high-security dialogue systems.
• Version Status: Includes Claude 3.5 – Sonnet version, with previous versions like Claude 2.

4. PaLM – 3 (Google)

• Developer: Google
• Features: Parameter scale exceeds 1 trillion, focuses on common sense reasoning and mathematical coding, leading response speed among similar models, supports 4096 tokens context.
• Application Scenarios: Automatic problem solving in education, financial quantitative model development.
• Version Status: Developed from the PaLM series to PaLM – 3 version, may have different fine-tuned versions.

5. LLaMA – 3 (Meta)

• Developer: Meta
• Features: Open-source model with 70 billion parameters, reasoning speed improved by 200%, performance close to GPT-4 in the open-source community, supports multilingual optimization.
• Application Scenarios: Customized AI solutions for small and medium enterprises, academic research.
• Version Status: Developed from the LLaMA series to LLaMA – 3 version, with community-driven secondary development versions.

6. Falcon – 200B (UAE TII)

• Developer: UAE Technology Innovation Institute
• Features: 180 billion parameter open-source model, mathematical reasoning and code generation capabilities on par with GPT-4, training costs only one-third of similar models.
• Application Scenarios: Multilingual services in the Middle East, low-cost AI infrastructure development.
• Version Status: Currently focused on Falcon – 200B version, with potential optimized versions in the future.

7. Cohere Command – R (Cohere)

• Developer: Cohere (founded by former Google team)
• Features: Focused on enterprise-level generative AI, supports 52 billion parameter scale, provides customized data privacy protection solutions.
• Application Scenarios: Customer service automation, intelligent management of internal documents.
• Version Status: Continuously iterates versions to meet various enterprise-level needs.

8. MPT – 50B (MosaicML)

• Developer: MosaicML
• Features: Open-source model with 8K tokens context length, lowest training costs in the industry, suitable for rapid deployment by small teams.
• Application Scenarios: MVP development for startups, experimental platforms for educational institutions.
• Version Status: Includes MPT – 50B version, with potential optimized versions for different application scenarios.

9. Nemotron – 4 (Nvidia)

• Developer: Nvidia
• Features: Integrates the Megatron framework, optimizes GPU computing efficiency, designed for AI chips, supports large-scale distributed training.
• Application Scenarios: Supercomputing centers, autonomous driving model training.
• Version Status: Continuously updates to adapt to new hardware and application needs.

10. Gopher – 2 (DeepMind)

• Developer: DeepMind
• Features: Reinforcement learning optimized version, sets records in game AI and protein structure prediction, supports multi-agent collaboration.
• Application Scenarios: Biomedicine research, complex game environment simulation.
• Version Status: Developed from the Gopher series to Gopher – 2 version, with potential fine-tuned versions for different fields.

Conclusion

This article presents the rankings of AI models in 2025, highlighting domestic models such as Tongyi Qianwen and Doubao Model, each with core capabilities and application scenarios, continuously updated. International models like GPT-4o and Gemini 2.0 Ultra also showcase unique features such as multimodal input and large-scale parameters. For detailed parameter comparison data of various AI models, click to view the comprehensive comparison metrics provided by Mijian.