Skip to content

Inference Platforms

If you sign up directly with OpenAI, Anthropic, or Google, you’re locked into their ecosystem:

“Don’t sign up for Google AI stuff or ChatGPT AI stuff or Anthropic AI stuff. Because AI services are going to be buying on the open market for a very low price.”

Direct subscriptions mean:

  • You can only use their models
  • You pay their prices (often higher)
  • Switching costs are high
  • You miss out on competition

Open Router provides access to 635+ models through a single API:

Terminal window
# Same API, any model
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-d '{
"model": "minimax/minimax-01",
"messages": [{"role": "user", "content": "Hello"}]
}'
  1. Single billing - One account, all models
  2. Easy switching - Change model name, that’s it
  3. Transparent pricing - 5% markup, that’s all
  4. Competition - Route to cheapest provider automatically

Open Router can integrate with external observability tools (request logging/monitoring) if you need it.

“You can connect external observability platforms into Open Router.”

This is useful when you operate a shared team setup and want auditing, cost monitoring, or debugging across many models.

Use models.dev to compare pricing across models. It’s maintained by the Open Router team and shows real-time prices.

New models often launch with free tiers to gather training data:

“New models, as they come out, they need more training data… they give it to inference platforms for one or two weeks for free.”

This means you can often use cutting-edge models for free if you’re willing to experiment.

Some interfaces also run their own free promotions via their preferred inference backends. For example, Open Code has an inference option called Open Code Zen.

Some platforms offer “anonymous” models:

“Big Pickle is a name for an unknown model. If you’re feeling adventurous, you can try it, and then maybe next week it’s going to turn out it was Facebook’s newest open source model.”

These are often pre-release models being tested. Fun for experimentation, not for production.

  1. Get an Open Router account - It’s the most flexible option
  2. Set a budget - Easy to track spending
  3. Default to cheap models - MiniMax, Kimi for daily work
  4. Route to expensive models only for complex tasks

You will hear a few overlapping words:

  • Inference platform/provider: runs models and serves an API (OpenAI, Anthropic, Open Router providers, etc.)
  • Router/aggregator: a layer that gives you one API and routes to many providers (Open Router)
  • Harness: industry slang for the UI/tool that wraps chat + tool calls (see the interfaces page)