Stack
OpenRouter
A router exposing 400+ models from 70+ providers behind one API key at pass-through prices, funded by a fee on credit top-ups.
What it actually does
OpenRouter puts 400+ models from 70+ providers behind a single API key, so you can swap GPT for Claude for DeepSeek by changing a model string. Inference itself is passed through at each provider's price with no markup; OpenRouter makes money on a fee applied to credit purchases.
It suits founders still comparing models or apps that let users pick their own model. A free-model catalog lets you benchmark at $0, capped at 50 requests/day without credits and 1,000 requests/day once you have bought $10+ in credits.
The costs are a 5.5% fee ($0.80 minimum) on every Stripe credit purchase (5% for crypto) and a BYOK charge of 5% of the normal model price after the first 1M requests/month. Once you settle on one model long-term, going direct to the provider saves the fee.
OpenRouter vs its main rivals
The tools people actually weigh against OpenRouter: Together AI, Vercel AI Gateway, Portkey. Same criteria for every column, including where OpenRouter loses.
| OpenRouter |  Together AI |  Vercel AI Gateway | PPortkey | |
|---|---|---|---|---|
Pricing & access | ||||
| Free tier | YesFree models, 50/day | No$5 min, no credits | YesFree tier credits | YesFree dev plan |
| Starting price | Pass-through + 5.5% | $0.17 /1M in | Pass-through | Free to $49/mo |
Models & capability | ||||
| Number of models | 400+ | 200+ (open) | 100s | 250+ |
| Frontier closed models included | YesYes | NoOpen only | YesYes | YesYes |
| Auto provider fallback | YesYes | NoSingle host | YesYes | YesYes |
Scaling & limits | ||||
| No markup on inference | No5.5% on top-ups | YesDirect host | Yes0% markup | YesSeat-priced |
| Bring-your-own-key | Yes5% after 1M/mo | NoNo | YesYes | YesYes |
Developer & API | ||||
| REST API access | Free | Paid | Free | Free |
| OpenAI-compatible endpoint | YesYes | YesYes | YesYes | YesYes |
Team & enterprise | ||||
| Built-in observability / analytics | YesBasic | YesBasic | YesYes | YesDeep (its focus) |
| Guardrails / governance | NoMinimal | NoMinimal | YesSome | YesYes |
- One API key for 400+ models from 70+ providers, swap models by changing a string
- No markup on inference itself; model prices are passed through from providers
- Free-model catalog is a zero-cost way to benchmark which model your feature actually needs
- 5.5% fee ($0.80 minimum) on every Stripe credit purchase is a pure tax versus going direct
- Free models cap at 50 requests/day across all free models combined, useless for production
- Thinner governance and observability than a dedicated gateway like Portkey
Pay less for it
5 ways foundThe 5.5% fee has a $0.80 minimum, so fewer larger top-ups reduce the per-dollar tax
Benchmark models at $0 before committing spend to any one
Use your own provider keys; the BYOK charge only starts after 1M requests/month
After settling on one model long-term, calling the provider directly skips the fee entirely
Vercel AI Gateway — Similar multi-provider routing and fallback with 0% inference markup, avoiding OpenRouter's credit-purchase fee once you no longer need its 400+ catalog
StackTracker tracks what you actually pay for OpenRouter and every other tool, flags overpayment, and shows the dollars you would save by switching.
Is it the right tool for you
- You are still comparing models and want them all behind one key
- Your app lets users choose their own model
- You want automatic provider fallback across closed and open models
- You have committed to one model long-term → use The provider's direct API to skip the 5.5% fee
- You need deep observability and governance for a team → use Portkey
Track what OpenRouter and the rest of your stack cost
StackTracker adds up every subscription, plus your hours, so you see the real number.
Prices and limits last verified 2026-07-20.