OpenRouter API (OpenRouter models): pricing, free tier, rate limits and how to start

Evaluate OpenRouter as an API routing and billing service with a model catalog behind it. Keep the selected model, serving endpoint and routing rules visible when comparing an integration.

What OpenRouter offers through the API

OpenRouter exposes a unified model API with native client SDKs and an OpenAI-compatible integration path. Official documentation.

Its role differs from a single model developer’s direct endpoint. Your application authenticates to OpenRouter and selects a catalog route, while a serving provider handles the model work. Keep those identities separate in logs and evaluation records. A familiar model name does not explain the complete path through which a request was served.

Provider routing can select among eligible endpoints, while an ordered model fallback list can try a different model when the primary route fails. Official documentation.

This flexibility is useful when the application needs to compare candidates or operate with a defined fallback. It also creates a design obligation: decide which changes are acceptable. A fallback that returns fluent text still needs to satisfy the same output contract, data requirements and cost expectations as the primary candidate.

Build an evaluation that records the actual result path. Retain the requested identifier, returned model, relevant routing configuration and accepted output. If the response quality changes, this evidence helps distinguish a prompt revision from a changed model or endpoint. Without it, the application can appear nondeterministic for reasons that are partly configuration changes.

OpenRouter’s model schema exposes identifiers, capabilities, supported parameters and pricing categories, including conditional overrides where applicable. Official documentation.

Use that metadata to create a shortlist, then test the task itself. A model that supports the required input type is eligible for evaluation, but capability metadata does not establish answer quality. Keep a safe fixture with a verifiable result and add a case that should fail gracefully. This prevents the routing layer’s convenience from becoming a substitute for application acceptance.

Account controls also belong in the design. Decide which workspace and key own the workload, which budget boundary should apply and what the application does when no eligible route remains. A router cannot guarantee that the intersection of every requested capability, price ceiling and account rule will always have an available endpoint.

The guides below separate these decisions. Pricing explains how to preserve billable categories and route assumptions. Models explains catalog identities and aliases. The free-tier page distinguishes a specific free variant from the free router. The key guide and tutorial create a small reproducible call, while Limits and Errors show how to investigate failures at the account, routing and upstream stages.

Start with one well-understood request and a restricted set of options. Add routing flexibility after the baseline works and after you have defined acceptable variation. This lets the application use OpenRouter’s catalog without losing the ability to explain which service produced a result or why a fallback was selected.

Pricing at a glance

Read OpenRouter pricing and route assumptions.

Models available

Use the catalog guide to distinguish model slugs, variants and aliases.

Free tier and how to get a key

Check free-model conditions and create an application credential.

Rate limits and tiers

Inspect account, key and endpoint limits before enabling parallel work.

Service status

Start in 5 minutes

Use the OpenRouter Python tutorial and retain the resolved model it prints.

client = OpenAI(api_key=os.environ["OPENROUTER_API_KEY"], base_url="https://openrouter.ai/api/v1")
result = client.chat.completions.create(model="~openai/gpt-sol-latest", messages=[{"role":"user","content":"Hello"}])
print(result.model)

Recent changes

Use the AI API cost calculator to turn the model and workload you are considering into an estimate.

Last verified 12 Sep 2026 · Source ↗

Frequently asked questions

Sources

• OpenRouter quickstart ↗
• Models and pricing schema ↗
• Authentication ↗
• Credit and rate limits ↗
• Errors and debugging ↗
• Provider routing ↗
• Free model variants ↗
• Free router ↗
• Workspace budgets ↗
• Prompt caching ↗
• BYOK ↗
• Provider logging ↗
• Batch quickstart ↗
• Latest aliases ↗

Last verified 12 Sep 2026 · Source ↗