One OpenAI-compatible API for DeepSeek, Qwen, and Chinese image/video models
NovAI has released an OpenAI‑style endpoint (https://aiapi-pro.com/v1) that routes requests to a suite of Chinese frontier models. By swapping the base_url in existing OpenAI SDK code, developers can invoke DeepSeek‑V4‑Pro for reasoning, Qwen‑3.7‑Max for multilingual chat, Doubao‑Seedream 5.0 for 2048×2048 image generation, and Doubao‑Seedance 2.0 for short video clips, all with the same authentication token. The video workflow uses a job‑submit and polling pattern, while chat and image calls follow the familiar synchronous or streaming patterns of the OpenAI API. Pricing claims show Qwen‑3.7‑Max about 45 % cheaper and Doubao‑Seed‑2.0‑Lite roughly 53 % cheaper than comparable offerings on OpenRouter, with DeepSeek‑V4‑Pro priced near parity.
This move reflects a broader push by Chinese AI labs to lower integration friction for global developers accustomed to OpenAI’s de‑facto standard. By exposing their models through a unified, OpenAI‑compatible layer, labs such as DeepSeek, Alibaba’s Doubao, and Meta‑like Qwen can tap into the massive ecosystem of tools, libraries, and codebases built around the openai Python client and similar SDKs. The approach also sidesteps the need for developers to juggle multiple SDKs, auth schemes, and billing consoles, a pain point that has limited cross‑model experimentation. Aggregators like OpenRouter have begun offering multi‑model access, but they typically focus on text‑only services; NovAI’s inclusion of image and video generation fills a notable gap.
The practical benefit is clear, yet developers should monitor a few risk factors. First, reliance on a third‑party gateway introduces an extra layer of latency and potential points of failure; service outages or policy changes at NovAI could disrupt access to all linked models. Second, the advertised cost advantages are based on published rates that can shift quickly, especially as Chinese providers adjust pricing to compete globally. Third, data residency and compliance concerns may arise when sending prompts to models hosted in China, especially for enterprises handling sensitive information. Finally, the video generation workflow’s polling loop may affect throughput and cost calculations, so developers need to benchmark end‑to‑end latency for production workloads.
Key Takeaways
NovAI’s OpenAI‑compatible endpoint consolidates access to Chinese LLM, image, and video models under a single API key and base URL.
Pricing for Qwen‑3.7‑Max and Doubao‑Seed‑2.0‑Lite is advertised as roughly half the cost of comparable OpenRouter rates.
The unified gateway eliminates the need for multiple SDKs and billing dashboards, streamlining multi‑modal development.
Dependence on a third‑party gateway adds latency, potential service risk, and compliance considerations that users must evaluate.
About the Source
This analysis is based on reporting by Dev.to Python. Here is a short excerpt for context:
If you build with LLMs, you already have OpenAI-compatible code lying around. The nice thing about...Read the original at Dev.to Python
