Most "free AI API" articles list trials that expire in 24 hours or caps of a few dozen requests. This is the short list of what is actually free in late August 2026 — verified today, no card required for any of it.
Z.AI's glm-4.6v-flash is served at $0 per token with no daily quota — a real vision-language model you can point at screenshots, invoices and product photos all day. For document parsing and image QA pipelines, this is currently the best $0 endpoint on the market.
from openai import OpenAI
client = OpenAI(api_key="sk-novai-...", base_url="https://aiapi-pro.com/v1")
r = client.chat.completions.create(
model="glm-4.6v-flash",
messages=[{"role": "user", "content": [
{"type": "text", "text": "Extract the total and date from this invoice."},
{"type": "image_url", "image_url": {"url": "https://example.com/inv.png"}},
]}],
)
Every new NovAI account gets $2.00 of free credit (14 days), no card. What that actually buys:
| Workload | Cost | Runs on $2 |
|---|---|---|
| qwen3.5-flash chat (800 in / 300 out) | ~$0.00015/call | ~13,000 calls |
| deepseek-v4-pro (same shape) | ~$0.0011/call | ~1,800 calls |
| Seedance video, 480p 5s | $0.70/clip | 2 clips |
Trial accounts can render one 480p/5s video per day; a first top-up of $1+ removes the cap and unlocks 720p and 15-second clips.
When the credit runs out, the floor price for a production-grade text model is $0.063 per million input tokens (qwen3.5-flash) — a month of hobby-project traffic typically costs under a dollar. Full numbers are in our cheapest-LLM ranking.
The honest summary: truly-free, unlimited AI APIs only exist at the small-model tier — but $2 of credit goes further on these price lists than most people expect.
Related: Paying for AI APIs without a credit card · Cheapest paid APIs