NVIDIA Opens Free APIs for Four Top Chinese AI Models

Overview

NVIDIA recently announced in its official API directory the formal opening of free calling interfaces for four top Chinese artificial intelligence models. This move marks further deep integration between the global AI ecosystem and Chinese large model technology, while also lowering the threshold for developers and enterprises to access high-end models.

The models opened this time include DeepSeek V4.1 Flash, GLM 5.3, GLM 5.3 Flash, and Kimi K3. Users can directly connect through NVIDIA’s unified API endpoint without additional complex configuration, enabling them to experience the capabilities of these models representing the current highest level of China's AI.

Core Model Technical Parameters

The four models included in the free interface all possess powerful parameter scales and architectural advantages, with particularly outstanding performance in the application of Mixture of Experts (MoE) technology.

• DeepSeek V4.1 Flash: As the latest evolved version of the DeepSeek series, the Flash optimized version maintains high inference speed while significantly reducing latency, making it suitable for application scenarios with high real-time requirements.
• GLM 5.3: Based on a 753B parameter mixture-of-experts architecture, it has extremely strong comprehensive understanding and generation capabilities, suitable for processing complex long-text analysis and professional domain tasks.
• GLM 5.3 Flash: A lightweight version of GLM 5.3, achieving a better balance between performance and efficiency, applicable to large-scale concurrent scenarios sensitive to costs.
• Kimi K3: Uses a super-large MoE architecture of approximately 2.8T scale, with excellent long-context processing capabilities, supporting deep understanding and information retrieval of ultra-long documents.

"NVIDIA, by opening these top-tier models, is accelerating the popularization and application implementation of China's AI technology within the global developer community." — Industry analysis perspective

Access Methods and Development Convenience

NVIDIA provides a unified API access address: https://integrate.api.nvidia.com/v1. Developers can call the above models through standard HTTP requests without separately applying for API permissions from each vendor, greatly simplifying the workflow of multi-model comparison testing and integration.

This move not only enhances the global accessibility of Chinese AI models but also reflects NVIDIA's strategic intent in building open and diversified AI infrastructure. As domestic model performance continues to break through, seamless docking with international platforms will further promote the two-way cycle of technological innovation and industrial application.