Limits

The values below are those in effect on November 5, 2026. The portal’s “Limits” tab shows the ones that apply to your workspace and lets you set spend caps (open).

Rate limits#Link to the Rate limits section

• The rate limit counts per key, not per workspace: several keys each have their own.
• A valid key that has just been used is not slowed down because someone else at the same address (corporate network, NAT) is using the wrong key.
• An IPv6 address is counted together with its whole /64 prefix.
• The portal’s “Limits” tab shows the per-key limit, under “Rate limits”.

Concurrent requests#Link to the Concurrent requests section

Chat has no limit on the number of concurrent requests: each request holds its share of the balance (see Hold), and it is the balance that limits what can run at the same time. When the holds in progress block a new request that would otherwise fit, the API responds with 429 credits_held_in_progress (Retry-After: 2).

Media endpoints have a number of requests in flight per workspace:

Beyond that: 429 too_many_concurrent with Retry-After (30 seconds for videos, 5 seconds otherwise).

Sizes#Link to the Sizes section

A multipart upload must include a Content-Length header (otherwise 411 length_required).

Timeouts#Link to the Timeouts section

Spend caps#Link to the Spend caps section

Two levels of caps limit what the API can spend. They are set in euros.

Level	Where	Windows
Workspace	“Limits” tab, “Workspace spend caps” section: “Per hour (€)”, “Per day (€)”, “Per month (€)”, then “Save limits”.	Sliding: the last hour, the last 24 hours and the last 30 days. Leave empty for no limit.
Key	When the key is created: “Spend cap (€)” and “Period”. See Authentication.	A sliding window of one hour, 24 hours or 30 days.

• Each request must fit within every cap that applies to it, hold included. Beyond that: 402 spend_cap_exceeded (with scope and window), or 402 hold_exceeds_cap when the hold of this single request exceeds what the cap leaves.
• The owner and the billing contact get an email when a cap is reached.
• A sliding window frees up over time: spend from 61 minutes ago no longer counts toward the hourly cap.

Retry without paying twice#Link to the Retry without paying twice section

Each HTTP request received has its own ID (X-Moon-Request-Id) and is billed only once. But the API does not recognize a retry on your side: sending the same request again creates a second one, with its own hold and its own cost. There is no idempotency header.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.realmoon.ai/v1",
    api_key=os.environ["MOONAI_API_KEY"],
    max_retries=0,
    timeout=600,
)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.realmoon.ai/v1',
  apiKey: process.env.MOONAI_API_KEY,
  maxRetries: 0,
  timeout: 600_000,
});