One model ID.
Automatic routing.

Three models. One chat model ID.

phantom-1 chooses from an economy, standard or frontier chat model at the start of a conversation. The selected model handles follow-ups unless the route is escalated or a fallback occurs. You pay the applicable rate of the model that actually runs.

How routing works

Phantom makes a model selection when a new chat conversation begins. It filters out the models that can’t accept the request, then selects from the models that remain. Follow-up requests stay with that model unless you ask to escalate or a fallback is needed.

A chat request with an image runs on an image-capable model outside the three text tiers.

curl https://phantom.codes/v1/key/route/test \
  -H "Authorization: Bearer $PHANTOM_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "phantom-1",
    "messages": [{
      "role": "user",
      "content": "Find the race condition in the webhook worker."
    }]
  }'

Control how requests are routed

Choose the model for each tier, when to skip the economy model, which providers may handle your data, or when to escalate a conversation. These controls affect which models Phantom can select. Phantom keeps no prompt or completion text; these flags govern what the upstream provider may keep, and they are off unless you send them.

curl -i https://phantom.codes/v1/chat/completions \
  -H "Authorization: Bearer $PHANTOM_KEY" \
  -H "Content-Type: application/json" \
  -H "x-phantom-session: refund-bug" \
  -d '{
    "model": "phantom-1",
    "zdr": true,
    "messages": [{
      "role": "user",
      "content": "Find the race condition in the webhook worker."
    }]
  }'

Request controls

Response metadata

HTTP/2 200
x-phantom-router: phantom-1
x-phantom-fallback: false
x-phantom-cost-usd: 0.003120
x-phantom-receipt: <signed receipt, base64>

Benchmarks

The last benchmark run, on October 25, 2026, tested today's router and models on 295 tasks. See the benchmarks page for its results.

One familiar model ID across API endpoints

Chat uses automatic routing. On supported non-chat endpoints, phantom-1 maps to a specific model for that operation. The same API key and model name work across these endpoints, but non-chat calls do not run the three-model chat router.