The head-to-head table
Every figure below is the vendor's own published number, re-checked against its pricing page on 29 September 2026 — these offers move, so treat the links at the bottom as the record.
| Gateway | What the free tier actually gives you | Fee on paid usage | Where it runs |
|---|---|---|---|
| FreeLLMAPI | 600+ models across 34 providers; ceiling is the sum of every provider free tier | None — the router is free, open source and self-hosted | On your machine; provider keys never leave it |
| OpenRouter | 25+ free models from 4 free providers, 50 requests/day (1,000 after a $10 spend) | 5.5% platform fee on paid credits (8% on Business) | Hosted; bring-your-own-keys on paid plans (keys stored with OpenRouter) |
| Requesty | All free models, 200 requests/day, no credit card | 5% markup on model cost once you pay | Hosted; bring-your-own-keys supported (keys stored with Requesty) |
| LLM Gateway | 3 free models at 20 requests/min; BYOK carries no platform fee | 5% fee on credit usage | Hosted, or self-hosted (AGPLv3) |
| Bytez | $1 free credits; open models capped at 7B params, 1 request at a time | Provider price + 2% on closed models | Hosted; keys live with Bytez |
What the table actually says
- Hosted free tiers are capped per account. OpenRouter's free plan is 25+ free models from 4 providers under a 50-requests/day cap (1,000 once you have bought $10 of credits). Requesty's free plan is all of its free models at 200 requests/day. LLM Gateway gives you 3 free models rate-limited to 20 requests/minute. Bytez hands you $1 in credits, with free-plan open models capped at 7B parameters and one concurrent request.
- "Free gateway" and "free on paid usage" are different promises. Every hosted gateway on this list takes a percentage the moment you spend money: 5.5% at OpenRouter, 5% at Requesty and LLM Gateway, 2% at Bytez. FreeLLMAPI has no usage to charge for — it is free, open-source software that runs on your machine and only ever calls provider free tiers.
- The keys question splits the field in two. With a hosted gateway, your requests transit their servers, and any keys you bring are stored there. With FreeLLMAPI you add each provider's free key once to your own router and nothing leaves the machine it runs on.
The structural difference
A hosted gateway's free models are a rate-limited allocation per account: OpenRouter's own docs cap its free model variants at 20 requests a minute and 50 a day, LLM Gateway caps its free models at 20 requests a minute, and Requesty at 200 requests a day. That is a fine deal when you want one bill, a dashboard and paid frontier models next to the free ones. It is a worse deal when all you want is the maximum free throughput your own keys can reach, because the gateway's cap applies on top of whatever the underlying provider allows, and your traffic runs through someone else's server.
FreeLLMAPI is the other shape of the same idea: an open-source router you run yourself that puts 600+ models across 34 providers behind one OpenAI-compatible endpoint — over 7.4 billion free tokens a month pooled across provider tiers — with automatic failover when one provider rate-limits, and no gateway cap, fee, or middleman holding your keys. The only paid thing is an optional catalog subscription ($19/year or $49 once), and routing is free without it.
Which one to pick
- Paid frontier models on one bill, plus some free models — OpenRouter or Requesty, accepting the daily cap and the platform fee.
- An open-source hosted gateway with BYOK — LLM Gateway, or self-host its AGPLv3 code.
- Bytez if you specifically want its 100,000+ model catalog and $1 of starter credits.
- Maximum genuinely free inference, keys kept private — FreeLLMAPI, self-hosted, across every provider's free tier at once.
Frequently asked questions
Which free LLM API gateway gives the most?
Measured as requests per day, the published caps are OpenRouter's 50/day, Requesty's 200/day, and LLM Gateway's 3 free models at 20 requests/minute. FreeLLMAPI is not capped by a gateway at all: its ceiling is the sum of the provider free tiers you hold keys for, currently 600+ models across 34 providers.
Do free gateways take a cut when you start paying?
Usually. OpenRouter charges a 5.5% platform fee on paid credits, Requesty a 5% markup on model cost, LLM Gateway a 5% fee on credits (though none on bring-your-own-keys traffic), and Bytez adds 2% on closed models. FreeLLMAPI takes nothing — the router is free software you host, and it only ever calls free tiers.
Are a hosted gateway's free models the same as provider free tiers?
Not quite. A hosted gateway's free models run under the gateway's own per-account limits (OpenRouter documents 20 requests/minute and 50/day on its free variants), and your requests pass through the gateway's servers. A self-hosted router calls each provider's free tier directly with your own key, under that provider's limits only, and nothing leaves your machine.
Can you run more than one gateway?
Yes, and gateways stack: a hosted gateway's cap is per account there, a self-hosted router's ceiling is per machine you run it on. In practice people use a hosted gateway when they want paid frontier models on one bill, and FreeLLMAPI when they want $0 inference across every provider free tier.
Figures checked 29 September 2026 against openrouter.ai/pricing, requesty.ai/pricing, llmgateway.io/pricing and docs.bytez.com. Free tiers and fees change often; check the source before relying on any row.
FreeLLMAPI vs OpenRouter → · FreeLLMAPI vs LiteLLM · Best free LLM APIs 2026 · Free LLM API: what is genuinely free