DEV Community

FastAnchor_io
FastAnchor_io

Posted on

I Built a Zero-Markup AI API Gateway - 18 Models at Provider Cost

The Problem

Every AI API gateway I tried added an invisible margin. OpenRouter, the biggest player, quietly marks up every model. You pay more than the provider charges, and you don't even know how much.

The Solution

I built aipossword.cn — an open-source AI API gateway with zero markup pricing.

  • 18 models across DeepSeek, Claude, Qwen
  • Zero markup — you pay exactly what providers charge
  • One endpoint — OpenAI compatible, one line code change
  • Open source — AGPLv3, built on New API (18k+ GitHub stars)
  • Free $5 credits on signup

Model Pricing

Model Input/1M Output/1M
DeepSeek V4 Pro $0.20 $0.87
DeepSeek V4 Flash $0.10 $0.20
Claude Opus 4.6 $15.00 $75.00
Claude Opus 4.7 $15.00 $75.00
Qwen 3.7 Max $1.25 $3.75

Quick Start

curl https://api.aipossword.cn/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"Hello"}]}'
Enter fullscreen mode Exit fullscreen mode

Why I Built This

I got tired of paying hidden fees on every API call. So I forked New API (18k stars), connected 18 models, and set markup to zero.

Is zero markup sustainable? I think the answer is yes — if enterprise users pay for SSO, SLA, and dedicated infrastructure. Individual developers get it free, at cost.

Try It

Visit aipossword.cn — free $5 credits, no credit card required.

Source: github.com/QuantumNous/new-api

Top comments (1)

Collapse
 
fastanchor_io profile image
FastAnchor_io

Love this concept—zero markup with 18 integrated models removes a huge layer of opaque cost overhead that third-party gateways usually pile on.
It’s way simpler to audit true API spending when there’s no extra vendor fee baked into every request. We’ve found tiered routing between cheap small models and heavy large models delivers the biggest savings, and your zero-margin setup makes that cost comparison crystal clear at a glance.
Curious how you handle baseline recalibration whenever users tweak routing configs or switch model weights? That silent cost shift is a common blind spot many gateway builders overlook.