OpenRouter :free limits (20 a minute, 50 a day) vs the Sume Free plan

On OpenRouter, a model id ending in :free is limited to 20 requests per minute, and to 50 requests per day until you have purchased $10 in credits, after which the daily cap is 1,000. Sume has no free model variant, only a Free plan: one job at a time, five queued.

Those are different kinds of limit, so comparing them directly would mislead. Here is what each page says.

What are OpenRouter's free-model limits?

The OpenRouter limits page lists limits for variants ending in :free. The per-minute cap is 20 requests for everyone. The daily cap depends on all-time credits purchased: fewer than $10 gives 50 requests a day, $10 or more gives 1,000. The daily counter resets on UTC days, and the per-minute limit is not exposed through the API.

Two other facts from the same page matter. A negative account balance triggers 402 even for free models until you add credits. And a 429 can come from an OpenRouter platform limit or from the upstream provider, with X-RateLimit-* headers and Retry-After guidance. OpenRouter's pricing page lists a Free tier with "25+ free models" from "4 free providers" with basic features.

What does Sume's Free plan limit?

Sume limits generation work by plan rather than by model. The Free plan processes one paid generation job at a time, queues up to five more, and so accepts six in total. Its request budget is 120 writes and 4,800 reads per minute, with reads and writes counted separately. Unauthenticated requests are limited per client IP at the Free rate.

A Sume submit also needs spendable balance: it reserves the estimated cost and returns 402 insufficient_credits before provider work starts if it cannot. The docs tell Free users to wait for included Gen$ or submit a cheaper request. Sume's documentation does not describe a no-cost model variant or a per-day request counter.

Why can't I treat them as the same thing?

OpenRouter's free variants are a set of models that cost nothing per request but are throttled hard. Sume's Free plan lets you run any catalog model your balance can reserve, but only one at a time. A prototype that fires ten requests in a burst fares differently on each: on OpenRouter you may use up a daily allowance, on Sume the extra jobs wait in queued (a normal state, not an error) and a seventh job beyond the six accepted returns 429 queue_full.

Note that the OpenRouter limits page speaks about requests in general; it does not tell you which free models cover video or image generation, so check the model list before assuming it fits a media workload.

How do I code against Sume's Free limits?

Read generation_limits from the submit response and stop adding work when queue_capacity_remaining is zero. Use the ratelimit-remaining header for request budget, and back off on retry-after.

curl -s -i https://api.sume.com/v1/balance \
  -H "Authorization: Bearer $SUME_API_KEY" | grep -i -E 'ratelimit|HTTP/'
# ratelimit-limit / ratelimit-remaining / ratelimit-reset describe the
# read or write bucket this request spent from; a 429 names it in
# error.details.scope (read or write).

Which should you use for a trial?

For a text-model experiment where a small daily allowance is enough, OpenRouter's free variants fit. For a media pipeline where you want to confirm the whole submit, poll and download path, Sume's Free plan, with one job at a time, is enough to test the wiring. Neither is a production capacity plan. For the paid side of the same question, read the OpenRouter in-flight budget note and Sume generation admission.