A gateway paying ten percent back on open models has published its own margin
Straitly returns 5 percent of spend on closed-source tokens and 10 percent on open-source. The gap between those two numbers is the most useful thing on the page.
Straitly launched an inference gateway on September 9 that pays cashback on token usage. Five percent back on closed-source models, ten percent on open-source, across 177 models behind one OpenAI-compatible endpoint, prepaid from $5 with no subscription. The cashback lands in a separate wallet and can be spent as credits or withdrawn as money.
Straitly has launched an inference gateway that pays cashback on token usage.
Users receive 5% cashback on closed-source models and 10% on open-source ones.
> Straitly’s API allows access to 177 models.
> It features automatic failover between providers and prepaid billing, with no subscription.
> Cashback lands in a separate wallet and can be spent as credits or withdrawn.
Tokens are a new currency 🔥
Ignore the offer. Look at the ratio.
A router can only give back what it makes. Nobody rebates a loss. Paying double on open weights says, roughly, that the margin on open models is likely at least twice the margin on closed ones — which is what you would expect when open weights are served by a crowd of interchangeable providers competing on price, and closed models come from one seller with a published rate card and no room underneath it. Five percent on a frontier model is probably close to the whole spread (which would explain why it is not six).
So the cashback table is a margin table with the labels changed. We have not seen another gateway publish one. Maybe none of them wanted to.
The second detail worth noticing is on Straitly's own homepage, where it compares itself with OpenRouter.
On a $100 basis, Straitly cashback is 5 dollars and OpenRouter's standard card credit-purchase fee is $5.50. Both charge for usage. This compares cashback with purchase fees, not total model prices.
A company that sets a rival's card-processing fee against its own rebate, and then writes the caveat itself in the same block, is doing something more honest than most comparison tables and still doing a comparison you should not lean on. Model prices are where the money is (per million tokens, on 177 models, none of them listed side by side) and those are not being compared at all.
The withdrawable wallet is the part we would actually watch. Most credit schemes are deliberately not cash, because non-withdrawable credit is a discount and withdrawable credit is a liability sitting on a balance sheet. Straitly has chosen the expensive version. That is a real commitment, and an unusual one for a startup routing other people's models.
So what do you think a router's margin is, if it can hand back a tenth of it and still trade?
Our read is that this is acquisition spend rather than a business model, and that the ten percent is the number that moves first. We would expect the open-source rate to fall below ten within six months, or the closed rate to go to zero, and if both rates hold through March we will have underrated how much fat sits in open-model routing.
Sources
Show all 4 sources
A rumour is now a starting gun, and mathematics just fired the first one
OpenAI spent a nine-figure token budget on someone else's problem because it heard they were close. Terence Tao says this is resource extraction. We think he is right, and that the fix will come from contracts, not from labs behaving better.
Continue ↓