Jarbird
OpenRouter-compatible API routing each request to the cheapest qualifying provider, capped at OpenRouter's list price.
Documentation
Example request
POST /api/v1/chat/completions
model
anthropic/claude-sonnet-5.5
charge
$0.840 list $1.400
Compare API prices.
USD per 1M input tokens and per 1M output tokens at standard uncached rates.
| Model | Jarbird | OpenRouter | Direct API |
|---|---|---|---|
| Claude Sonnet 5.5 | 1M in $0.4641 1M out $2.3206 | 1M in $2.11 1M out $10.55 | 1M in $2.00 1M out $10.00 Anthropic |
| Claude Opus 5.5 | 1M in $0.9282 1M out $4.6411 | 1M in $4.22 1M out $21.10 | 1M in $4.00 1M out $20.00 Anthropic |
| Claude Fable 5.1 | 1M in $2.3206 1M out $11.6028 | 1M in $10.55 1M out $52.75 | 1M in $10.00 1M out $50.00 Anthropic |
| GPT-6.1 Sol | 1M in $1.298 1M out $6.49 | 1M in $2.11 1M out $10.55 | 1M in $2.00 1M out $10.00 OpenAI |
| GPT-6 Sol | 1M in $1.298 1M out $6.49 | 1M in $2.11 1M out $10.55 | 1M in $2.00 1M out $10.00 OpenAI |
| GPT-6 Luna | 1M in $0.0649 1M out $0.3245 | 1M in $0.1055 1M out $0.5275 | 1M in $0.10 1M out $0.50 OpenAI |
Jarbird shows one rate pair per model, selected on the lowest combined input and output rate. OpenRouter estimates include the 5.5% Standard credit-purchase fee, spread across usage.
Change one line.
You keep the OpenAI SDK and your OpenRouter model IDs. Point the base URL at Jarbird and use your Jarbird key. The requests themselves stay the same.
Python, OpenAI SDK
client = OpenAI(
-Removed: base_url="https://openrouter.ai/api/v1",
+Added: base_url="https://jarbird.com/api/v1",
api_key=os.environ["API_KEY"],
)
Six things decide each route.
-
Sensitive requests
Requests matching our secret checks go only to OpenRouter. -
Capabilities
A route has to support the model, tools, images and context length you asked for. -
Prompt caching
If a route already has the start of your prompt cached, that part costs the route's cache rate. -
Price, then reliability and speed
The cheapest qualifying route wins. If two prices are within 2%, the tie goes to the route with your prompt cached, then to fewer recent errors, then to faster first tokens. -
Quality checks
Providers are audited by model. A failed audit stops traffic for that model. -
Failover
If a route fails before internal commitment, the request can move to the next one, with up to three attempts total. Commitment can happen before the first upstream byte is forwarded.

A heron standing in a still mountain lake at night, a hut on the far shore.
Pay as you go.
Top up your prepaid balance with USDC or USDT. Failed attempts cost nothing.
Charges never exceed the equivalent OpenRouter list price.
Example request, anthropic/claude-sonnet-5.5
| OpenRouter list price | |
|---|---|
| You pay | $0.840 |
| Estimated savings | $0.560 |
These figures are an example. Real prices change with the providers.
Questions before you switch.
Which models can I use?
Check the model catalogue for API availability. Copy the exact model ID into the model field of your request.
Does Jarbird support streaming?
Yes. Jarbird can try up to three routes per request. It can switch routes only before it commits to a response. That can happen before your client receives any data. If Jarbird blocks the stream after that point, it closes the stream.
Where can I find the route and cost?
Check provider in the response for the route. usage.cost shows what Jarbird charged. usage.cost_details.baseline_cost shows the OpenRouter list price for the same request.
Which API features are unavailable?
Jarbird doesn’t currently support the Responses API, the Anthropic Messages API, embeddings, image generation, or API keys from other providers.
How do I get access?
Jarbird is in invite-only beta. Use the form below to request an invite.
Request an invite.
Briefly describe how you use language models and estimate your monthly API spend.
Use Jarbird only for lawful requests. If a provider flags requests made with your Jarbird API key, we suspend that key.
We use your email, use case, and spending estimate only to review and reply to your invite request.