Agnes AI Launches a China Node: One-Line Config Turns Hermes from Detour to Direct Connection, with a Free Multimodal API

For Hermes users, Agnes AI — a Singapore-based model — is well known because it offers fully free multimodal support for text, image generation, and video generation. Yesterday came the news that Agnes launched a domestic official domain (agnes-ai.cn) and simultaneously switched its API node to mainland China. Only the address needs to change; your existing API Key still works.

💡 What You Will Learn

For Hermes users, Agnes AI — a Singapore-based model — is well known because it offers fully free multimodal support for text, image generation, and video generation. Yesterday came the news that Agne

On July 29, Agnes AI officially enabled its domestic website agnes-ai.cn, along with a domestic API node. The API endpoint changed from apihub.agnes-ai.com/v1 to apihub.agnes-ai.cn/v1, and nothing else needs to change: the API Key, model names, request parameters, and calling method all stay the same. After replacing the address and restarting the application, you'll be routed to the domestic node.

The operating entity is Nanjing Aisi Artificial Intelligence Technology Co., Ltd., whose parent company Sapiens AI is headquartered in Singapore. This means Agnes now has a legal entity in China, providing a foundation for compliant operations, filing, and enterprise cooperation going forward.

Alongside the domestic site launch, the first-round evaluation results for Agnes 2.5 Pro Alpha on Artificial Analysis were also published. It scored 39 on the Intelligence Index, ranking 9th among 153 evaluated models. This index combines real professional tasks, agent tool use, terminal coding, scientific programming, complex reasoning, and long-context capabilities, making it closer to real-world performance than single knowledge-question ability. It scored 58.8 on the Coding Index and 67.0% on Terminal-Bench v2.1, which requires the model to enter a real terminal environment, execute commands, read and write files, and complete software engineering tasks — very close to how today's AI coding agents are actually used. Agnes 2.5 Pro Alpha also supports a 1M long-context window for handling larger code repositories and business materials.

Agnes 2.5 Flash targets everyday coding, agent, and high-frequency development tasks, suitable for document processing, code generation, bug fixing, and project modification. Crucially, it remains free indefinitely with no credit card or prepayment required, so everyday development scenarios can keep running at zero cost.

Agnes recently disclosed usage data: its multimodal models processed 8.16 trillion tokens in a single week, with 5.10 trillion (62.5%) text and 3.06 trillion (37.5%) multimodal. Text remains the mainstream, but image and video generation already contribute more than a third of token volume — the platform's usage is shifting toward multimodal. Beyond the models, the Agnes team is also contributing to the SGLang open-source community, with 4 pull requests merged covering multi-GPU parallelism, VRAM optimization, and KV-cache stability.

For developers, the domestic node's biggest benefit is lower latency — no more routing through international links. Previously API requests had to go overseas; now domestic direct connection means faster and more stable responses. For Hermes users, the configuration only requires changing one line of address — swapping the base_url — while the model names and API Key are all reused. More importantly, Agnes's free strategy hasn't changed with the domestic site launch: 2.5 Flash stays free, and 2.5 Pro Alpha has entered developer beta testing, with a full release likely not far off.