DEV Community

仪袁韶
仪袁韶

Posted on

The '6x Markup' Panic: What a Managed LLM Gateway Actually Sells You

#deepseek #api #pricing #llm

This is a follow-up to my field guide on using DeepSeek/GLM/Qwen from anywhere. Someone in the comments captured the mood perfectly: "a 6x markup makes zero sense." They're not wrong to feel that. But "markup" and "price" are two different numbers, and confusing them is how people pick the wrong tool. Let me untangle it — without the sales gloss.

First, the panic is half-right

If you measure a gateway only by "how many dollars per million tokens for DeepSeek-V4-Flash," then yes: a managed gateway that sources through a compliant retail channel will look 3–6× more expensive than a bare aggregator running on cheap wholesale compute.

That comparison is legitimate. It's also incomplete. Here's the part that gets skipped.

Two different products wear the same "API" label

When you buy tokens from a bare aggregator, you are buying one thing: inference, at their wholesale cost plus their cut.

When you buy from a managed gateway, you are buying a bundle:

  • one OpenAI-compatible endpoint for many Chinese models (DeepSeek, GLM, Qwen, Kimi…) instead of N separate integrations
  • one USD-denominated bill (every charge shown in USD — no RMB-conversion surprise at month-end) instead of N provider consoles
  • transparent pricing you can audit — at TideLink it's literally upstream cost × 1.30, no mystery tier
  • a real resp.model you can verify, so you know which model actually served your request
  • a free GLM tier (glm-4.7-flash) that needs no card and no +86, as a $0 on-ramp

None of that changes the token price. It changes what the token price is buying.

The honest verdict: for pure price, we are not it

I'll say it plainly so you can trust the rest: if your only criterion is the absolute lowest DeepSeek price-per-token, do not use a managed gateway. Use OpenRouter, Novita, or sign up to the source directly. A gateway's convenience costs money, and on flagship DeepSeek that premium is real (our margin is a flat 30% on upstream cost — the upstream cost itself is what's higher, because we source through compliant retail channels rather than wholesale tiers that forbid resale).

Pretending otherwise is how this whole market earns its "sketchy" reputation. So no: we are not cheapest. Stated.

So when is the premium worth it?

For the developer who hits the access-friction wall rather than the price wall:

  • You tried the official site and it wanted a +86 number.
  • You tried three providers and now have three dashboards, three invoices your accountant can't reconcile, three rate-limit policies.
  • You want to prototype this week without standing up a GPU box.

For that person, the question isn't "is $0.30/M cheaper than $0.54/M" — it's "how many hours of integration and billing plumbing am I trading for this premium?" Often the gateway wins on total effort even at a higher unit price. That's the actual product. Not the token.

The move I'd actually recommend

Don't argue with the markup. Decouple the two jobs:

  1. Prototype and do real work for free on glm-4.7-flash via a gateway. No card, no +86, real model, real resp.model. Ship something.
  2. When you genuinely need DeepSeek/flagship quality, pay the transparent upstream × 1.30 and see the number before you call.
  3. If you're at scale (millions of tokens/day) and price-dominated, self-host the open weights or negotiate wholesale — a gateway will never beat that on unit cost, and anyone claiming it will is lying.

That sequence gets you moving today at $0, keeps your exit open, and never asks you to pretend a convenience fee is free.

What we are, in one line

Not the cheapest. The lowest-friction honest on-ramp: free GLM entry, transparent cost × 1.30, a verifiable resp.model, USD-denominated billing, and enterprise invoicing / volume pricing available on request. If "access friction" is your real problem, that's the road.

Written by the team behind TideLink. USD billing and enterprise invoicing/volume pricing are available on request via support@tidelink.xyz — we'd rather be straight about what's live vs. negotiable than oversell. Attribution tag for our own analytics: ?cid=devto.

Get a free TideLink API key — call GLM, Qwen, DeepSeek and more through one OpenAI-compatible endpoint, no card required for the free GLM tier: https://tidelink.xyz/dashboard.html?cid=devto

Top comments (0)