AI & ML APIs

OpenRouter

A unified API gateway to 300+ AI models (Llama, Qwen, Gemini, Claude, and more) with free :free model variants — one key, one API, instant model switching.

Use Cases

A/B test five open models on the same prompt to pick the best one for your app before committing

Build a model-agnostic chatbot that switches between free and paid models based on traffic

Prototype an AI feature using free variants, then scale the same code to a paid model without rewrites

Free Tier

Free model variants (:free) at 20 requests/min and 50 requests/day (1,000/day after $10 lifetime credits), no credit card required

How to Maximize the Free Tier

The free tier routes you to :free model variants at 20 RPM and 50 requests/day — plenty for testing, and it jumps to 1,000/day once your account has purchased $10 in lifetime credits. Use OpenRouter as your model-testing lab: try the same prompt across Llama, Qwen, and Gemini via a single API before committing to a dedicated provider, and lock model routing with `provider` preferences to control which backend serves you. Because the 50/day cap is shared across all free models, cache repeated results server-side and batch experiments. Combine with a serverless cache (Upstash or Cloudflare Workers) to avoid burning the daily quota on identical calls. The real win: when you outgrow free, you already have one integration that can call any paid model — no migration.

Getting Started

Sign up at openrouter.ai (email or GitHub) → create an API key in the dashboard (no credit card) → call `https://openrouter.ai/api/v1/chat/completions` with a `:free` model ID like `meta-llama/llama-3.1-8b-instruct:free` → check remaining quota with `GET /api/v1/key` → switch models by changing one field in the request.

Pros

One API, all models: A single key and OpenAI-compatible endpoint reach 300+ models — swap providers by changing one string
Real free tier: :free model variants work with no credit card — genuinely usable for prototyping and evaluation
Model routing control: Fallback routing and provider preferences let you manage cost, latency, and reliability per request

Cons

Low free quota: 50 requests/day on free models exhausts quickly for anything beyond light testing
Free tier volatility: :free variants depend on provider availability and can be deprecated or rate-limited upstream at any time
Markup on paid: Paid models carry an OpenRouter fee on top of provider pricing — cheaper direct, but you pay for convenience

Alternatives

Hugging Face Inference API8/10★★★★☆Groq8/10★★★★☆DeepSeek8/10★★★★☆