AI生成 into the corner of the output under China’s labelling rules — NovAI cannot switch it off. Video output varies by model: in our 4 August 2026 tests Seedance 2.0 and CogVideoX-Flash showed no visible label, while Hunyuan Video 1.5 is labelled. Text and chat models are not affected. Which models, and what it rules out →Chat, vision reasoning, image generation and video generation — all at $0, all through one API key. We run NovAI, so this is a first-party announcement, not a review.
| Model ID | Type | Endpoint | Price |
|---|---|---|---|
glm-4.7-flash | Chat LLM | /v1/chat/completions | $0 |
glm-4.1v-thinking-flash | Vision + reasoning | /v1/chat/completions | $0 |
glm-4.6v-flash | Vision | /v1/chat/completions | $0 |
cogview-3-flash | Image generation | /v1/images/generations | $0 |
cogvideox-flash | Video generation | /v1/video/generations | $0 |
All five are Zhipu (Z.ai) GLM-family "flash" variants. Flash models are smaller and faster than flagship models — the trade-off for $0 is capability, not hidden fees.
Zhipu offers these flash models free upstream, and NovAI passes that through with zero markup. You still get the same OpenAI-compatible interface, English dashboard, and email signup (no Chinese phone number) as with paid models.
from openai import OpenAI
client = OpenAI(api_key="sk-novai-xxx", base_url="https://aiapi-pro.com/v1")
resp = client.chat.completions.create(
model="glm-4.7-flash",
messages=[{"role": "user", "content": "Summarize this in one sentence: ..."}],
)
print(resp.choices[0].message.content)resp = client.chat.completions.create(
model="glm-4.1v-thinking-flash",
messages=[{"role": "user", "content": [
{"type": "image_url", "image_url": {"url": "https://example.com/chart.png"}},
{"type": "text", "text": "What trend does this chart show?"},
]}],
)The "thinking" variant reasons step-by-step over images — useful for charts, documents and screenshots. For simpler image description tasks, glm-4.6v-flash uses the same request format.
img = client.images.generate(
model="cogview-3-flash",
prompt="minimalist logo of a paper plane, flat design",
)
print(img.data[0].url)import requests
r = requests.post(
"https://aiapi-pro.com/v1/video/generations",
headers={"Authorization": "Bearer sk-novai-xxx"},
json={"model": "cogvideox-flash", "prompt": "a paper plane flying over a city at sunset"},
)
print(r.json())Video generation is asynchronous — you receive a task ID and poll for the result.
When you outgrow the flash models, the same API key works for the paid catalog (DeepSeek, Qwen, Kimi, Doubao Seedance video, Seedream image and more) — see the model marketplace.
Sign up with email, grab an API key, and call all 5 free models today.
Five models: GLM-4.7-Flash (chat), GLM-4.1V-Thinking-Flash (vision reasoning), GLM-4.6V-Flash (vision), CogView-3-Flash (image generation) and CogVideoX-Flash (video generation).
Yes. Point your OpenAI SDK at https://aiapi-pro.com/v1 and use the free model IDs — no other code changes.
No. Email signup; the free models cost nothing, and paid models accept USDT (TRC20).
No hidden fees — the upstream provider (Zhipu) offers these flash models free and we pass that through. The trade-off is that flash models are smaller than flagship models.