DEV Community

Marcene
Marcene

Posted on

I built an AI API aggregator that saves developers 60-85% on model costs

I built an AI API aggregator that saves developers 60-85% on model costs

The Problem

I'm a solo developer who uses AI models daily — GPT-4o for complex reasoning, Claude for long documents, DeepSeek for coding, and MiniMax for image generation.

Managing 5 different API accounts was painful:

  • 5 different billing cycles
  • 5 different SDKs
  • 5 different rate limits to track

And the costs? OpenAI charges $2.50 per 1M input tokens for GPT-4o. Anthropic charges $3.00 for Claude Sonnet. If you use both regularly, your monthly bill hits triple digits fast.

The Solution: Celuxe

I built Celuxe — a unified API that aggregates 200+ AI models behind a single OpenAI-compatible endpoint.

Replace your OPENAI_BASE_URL and keep your existing code.

That's it. No new SDK. No migration. One line change.

Real Cost Comparison

Here's what I'm actually paying vs official pricing:

Model Official (per 1M tokens) Celuxe (per 1M tokens) Savings
GPT-4o $2.50 / $10.00 $0.80 / $3.20 68%
Claude 3.5 Sonnet $3.00 / $15.00 $1.20 / $6.00 60%
DeepSeek V3 $0.27 / $1.10 $0.14 / $0.55 50%
Gemini 2.0 Flash $0.15 / $0.60 $0.06 / $0.24 60%
MiniMax M2.7 $0.30 / $1.20 $0.15 / $0.60 50%
GPT-4o Mini $0.15 / $0.60 $0.06 / $0.24 60%
Claude 3.5 Haiku $0.80 / $4.00 $0.30 / $1.50 63%

Before Celuxe, my monthly AI bill was ~$200. Now it's ~$50. Same models. Same quality.

The 16 Free Dev Tools

While building the API, I realized developers need more than just cheap model access. So I built a free tools suite:

  • Model Compare — side-by-side pricing comparison
  • Cost Calculator — estimate your monthly bill before deploying
  • Playground — test any model without signing up (3 free trials)
  • Code Generator — cURL/Python/Node.js code snippets ready to copy
  • JSON Formatter, Base64, UUID Generator, Regex Tester, Markdown Preview, and 6 more utilities

All 16 tools are free, no login required.

Tech Stack

  • Backend: One API (Go) for model routing + load balancing
  • Frontend: Static HTML + Tailwind CDN (zero JS framework)
  • Infrastructure: US-based VPS, Cloudflare DNS, Nginx reverse proxy
  • Models: OpenAI, Anthropic, DeepSeek, Google, MiniMax — all through unified /v1/chat/completions

What I Learned

  1. Static sites scale. 50+ pages of pure HTML serve faster than any Next.js app
  2. Developer tools are the best SEO. JSON Formatter alone gets thousands of hits
  3. OpenAI compatibility is table stakes. If you're not a drop-in replacement, developers won't bother
  4. Transparency converts. Publishing real comparison data builds trust

Try It

Celuxe is live. New users get 500,000 free tokens — no credit card required.

GitHub: github.com/xiaojin/celuxe-sdk

If you're building with AI APIs, I'd love to hear about your cost challenges in the comments.

Top comments (0)