fal.ai Concurrency Limit Explained: Why Requests Queue and How to Get More Than 40
fal.ai limits new accounts to 2 concurrent requests and raises it automatically to at most 40 as you buy credits. How the limit works, how to tell queueing from slowness, and your options beyond 40.
fal.ai does not limit requests per minute. It limits how many of your requests run at the same time. A new account can run 2 requests at once, the limit rises automatically as you buy credits up to a maximum of 40, and anything above that is arranged through sales. Requests over the limit are not rejected: they wait in a queue, which is why the symptom is slow jobs, not errors.
This guide explains how to recognize the limit, how to work within it, and what to do when 40 is not enough.
Sources checked November 22, 2026: fal’s documentation and WaveSpeedAI’s documentation.
What fal’s rule says
From fal’s FAQ:
- New accounts start at 2 concurrent requests.
- “As you purchase credits, the limit increases automatically up to 40.”
- Requests beyond the limit wait in a queue until a slot is free.
- Higher limits are available through sales.
The FAQ does not publish the purchase amounts that correspond to each step between 2 and 40.
How to tell you are hitting it
Because extra requests queue instead of failing, the limit looks like a performance problem.
- Wall-clock time grows with batch size. Ten video jobs take about five times as long as two.
- Requests sit in the queue state before they start, and the wait is proportional to how many you submitted.
- Single requests are fast. If one request alone finishes quickly and a batch does not, it is the concurrency limit, not the model.
A quick test: submit one request and time it, then submit ten at once and watch how many are in progress at the same moment. That number is your limit.
What the limit means for a product
Concurrency translates directly into throughput. If a video takes two minutes to generate:
| Concurrent requests | Videos per hour |
|---|---|
| 2 | 60 |
| 10 | 300 |
| 40 | 1,200 |
For a product where users wait for a result, the more relevant number is how many users can be served at once. With a limit of 2, the third user waits for someone else’s job to finish.
Working within the limit
- Run your own queue. Submit at most your limit and hold the rest in your application, so you control priority. Paying users and interactive requests should go before batch jobs.
- Use webhooks instead of holding connections open. A queued request can wait a long time; do not tie up a worker for it.
- Separate interactive and batch traffic. Give batch work a lower share of the slots, or run it at night.
- Generate drafts on faster models. A shorter job frees its slot sooner.
- Show users their position. A queue with feedback feels much shorter than a spinner.
import asyncio
LIMIT = 2 # set this to your account's concurrency limit
slots = asyncio.Semaphore(LIMIT)
async def generate(job):
async with slots:
return await run_on_provider(job)
async def run_all(jobs):
# Interactive jobs first, batch jobs after.
ordered = sorted(jobs, key=lambda job: job.priority)
return await asyncio.gather(*(generate(job) for job in ordered))
When 40 is not enough
You have three options.
- Buy your way up to 40. The limit rises automatically with credit purchases. Remember that purchased credits on fal expire after 365 days, so do not top up more than you will use only to raise a limit.
- Talk to fal’s sales team for a higher limit. This suits teams with steady, predictable volume.
- Spread the load across two platforms. Route by model or by traffic type, so that each platform’s limit applies to only part of your traffic. This also protects you from an account-level problem on either one.
Concurrency on WaveSpeedAI
WaveSpeedAI publishes its limits by account level in its documentation:
| Level | Concurrent tasks | Predictions per minute | How you reach it |
|---|---|---|---|
| Bronze | 2 | 5 | Default for new accounts |
| Silver | 300 | 500 | Any successful top-up |
| Gold | 3,000 | 3,000 | A single top-up of $1,000 or more |
| Ultra | 10,000 | 5,000 | A single top-up of $5,000 or more |
A new account starts with the same 2 concurrent tasks. The difference is the next step: any top-up moves the account to 300 concurrent tasks, and the upgrade applies automatically. Limits above Ultra are available on request.
Two details to plan for:
- There is also a limit on predictions started per minute. A request over it returns
429, so use bounded backoff for submissions. - Levels are set by the size of a single top-up, not by cumulative spend.
The rest of the billing model is simple: credits never expire, failed tasks are refunded in full, and the price of a request is shown before you run it. The same model families are available, including Seedance, Kling, Veo, Wan, Nano Banana and Seedream.
To move an integration over, see how to migrate from fal.ai to WaveSpeedAI.
FAQ
What is the fal.ai rate limit?
fal limits concurrent requests, not requests per minute. New accounts can run 2 requests at once, and the limit rises automatically with credit purchases to a maximum of 40.
Why are my fal requests stuck in the queue?
You have more requests in flight than your concurrency limit allows. The extra ones wait until a running request finishes.
How do I increase my fal concurrency limit?
Buying credits raises it automatically, up to 40. Beyond that you need to contact fal’s sales team.
Does fal return an error when I exceed the limit?
No. Requests over the limit are queued, so the effect is waiting time, not a 429.
Which platform allows more concurrent video generations?
On WaveSpeedAI, any top-up raises an account to 300 concurrent tasks, with 3,000 and 10,000 at the higher levels. fal’s self-serve maximum is 40.
/filters:quality(82)/media/images/1773962750383987480_n3hqzHRZ.webp)