Model Router

Select a model

Classifies a task with Neon 1.1 and selects a model from the benchmark catalog. Requires runs_write.

Billed as one decision run: the same admission, quota, credit and settlement as POST /v1/runs with "kind": "decision". Send the task as prompt or as messages, not both. The body may be up to 4 MiB; decisions read up to 262,144 input tokens.

Idempotency-Key is optional. Replaying a key returns the stored classification with replayed: true and never charges twice.

POST

router

select

Select a model

curl --request POST \
  --url https://api.opentype.dev/v1/router/select \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{}'

const options = {
  method: 'POST',
  headers: {Authorization: 'Bearer <token>', 'Content-Type': 'application/json'},
  body: JSON.stringify({})
};

fetch('https://api.opentype.dev/v1/router/select', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

import requests

url = "https://api.opentype.dev/v1/router/select"

payload = {}
headers = {
    "Authorization": "Bearer <token>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

{
  "id": "rtr_0f8e3c1a9b2d4e5f8a7b6c5d4e3f2a1b",
  "run_id": "<string>",
  "policy": "<string>",
  "model": {
    "id": "gpt-6-sol",
    "name": "GPT-6 Sol",
    "provider": "openai",
    "open_weights": true
  },
  "classification": {
    "domain": {
      "label": "<string>",
      "probabilities": {}
    },
    "difficulty": {
      "label": "<string>",
      "probabilities": {}
    },
    "task_type": {
      "label": "code_generation",
      "family": "<string>",
      "top": [
        {
          "task_type": "code_generation",
          "family": "coding",
          "probability": 123
        }
      ],
      "fixed": true
    },
    "facets": {
      "difficulty_expected": 123,
      "difficulty": {
        "label": "<string>",
        "probabilities": {}
      },
      "output_length": {
        "label": "<string>",
        "probabilities": {}
      },
      "output_tokens_est": 123,
      "needs_tools": 123,
      "needs_vision": 123,
      "safety_sensitive": 123,
      "language": "<string>",
      "input_tokens_est": 1,
      "target_language": "<string>"
    }
  },
  "ranking": [
    {
      "id": "<string>",
      "name": "<string>",
      "provider": "<string>",
      "open_weights": true,
      "score": 123,
      "blended_price_per_mtok": 123,
      "domain_score": 123,
      "expected_quality": 123,
      "uncertainty": 123,
      "estimated_cost_usd": 123,
      "estimated_latency_ms": 123,
      "latency_estimated": true,
      "strengths": [
        {
          "benchmark": "swe_bench_verified",
          "name": "SWE-bench Verified",
          "norm": 123,
          "weight": 123,
          "contribution": 123,
          "value": 123,
          "rank": 1,
          "gap_to_best": 123
        }
      ],
      "weaknesses": [
        {
          "benchmark": "swe_bench_verified",
          "name": "SWE-bench Verified",
          "norm": 123,
          "weight": 123,
          "contribution": 123,
          "value": 123,
          "rank": 1,
          "gap_to_best": 123
        }
      ],
      "imputed": [
        {
          "benchmark": "<string>",
          "rule": "<string>",
          "from": "<string>"
        }
      ]
    }
  ],
  "score_basis": "domain",
  "filters_applied": [
    {
      "filter": "<string>",
      "removed": 1
    }
  ],
  "low_confidence": true,
  "input_tokens_est": 1,
  "reason": "<string>",
  "decision_model": "neon-1.1",
  "catalog_as_of": "<string>",
  "benchmarks_as_of": "<string>",
  "replayed": true,
  "difficulty_floor": 123,
  "threshold": {
    "q_star": 123,
    "r": 123,
    "tau": 123
  },
  "usage": {
    "input_tokens": 1,
    "output_tokens": 1
  },
  "cost_micros": 1
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

Authorizations

Authorization

string

header

required

An API key (otsk_ + 64 hex) or a console session token.

Headers

Idempotency-Key

string | null

Optional. Replaying a key returns the stored classification and never charges twice.

Body

application/json

A selection request: the task, as prompt or as messages (exactly one).

prompt

string | null

The task. Bodies up to 4 MiB are accepted; only the head and tail of the task are read to classify it.

messages

object[]

The task as conversation turns, joined into one text as for prompt.

Show child attributes

policy

string | null

balanced (default), cost_efficient, capability_heavy or domain_skills. Anything else is 400 invalid_policy.

Example:

"balanced"

domain

string | null

v1 override: route as this domain's default task type (coding, math, reasoning, knowledge, agentic, long_context, writing, multilingual or general). The task type is then not classified; difficulty and the facets still are.

task_type

string | null

Skip task-type classification and route as this task type (see GET /v1/router/task-types). Wins over domain.

Example:

"code_generation"

latency

string | null

interactive, standard (default) or batch: how much estimated latency weighs in balanced.

Example:

"standard"

max_latency_ms

number<double> | null

Drop models whose estimated time to the full answer exceeds this, or whose speed is not measured.

weights

null | object

balanced only: replaces the quality, cost and latency weights. With any other policy it is 400 weights_require_balanced.

Show child attributes

models

null | object

Filters over the catalog. All optional; an empty object keeps every model.

Show child attributes

Response

The selected model, the classification and the ranking

string

required

rtr_ plus the uuid of the classification run.

Example:

"rtr_0f8e3c1a9b2d4e5f8a7b6c5d4e3f2a1b"

run_id

string

required

The classification run, readable at /v1/runs/{run_id} and in usage.

policy

string

required

model

object

required

A model, as a selection names it.

Show child attributes

classification

object

required

Show child attributes

ranking

object[]

required

Top candidates, best first, at most ten.

Show child attributes

score_basis

enum<string>

required

Which benchmarks decided. v2 always decides on the task's weighted benchmark vector; domain and intelligence are the v1 values.

Available options:

domain,

intelligence,

benchmarks

filters_applied

object[]

required

Filters that removed at least one model, in the order applied.

Show child attributes

low_confidence

boolean

required

domain_skills with a most probable task type under 40%.

input_tokens_est

integer<int64>

required

Required range: x >= 0

reason

string

required

decision_model

string

required

Example:

"neon-1.1"

catalog_as_of

string

required

The catalog snapshot date.

benchmarks_as_of

string

required

The benchmark snapshot date.

replayed

boolean

required

True when an Idempotency-Key replayed a stored classification.

difficulty_floor

number<double> | null

v1 name for threshold.tau (balanced, cost_efficient).

threshold

null | object

The quality bar (balanced, cost_efficient).

Show child attributes

usage

null | object

Tokens the run used.

Show child attributes

cost_micros

integer<int64> | null

Required range: x >= 0

Select a model

curl --request POST \
  --url https://api.opentype.dev/v1/router/select \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{}'

const options = {
  method: 'POST',
  headers: {Authorization: 'Bearer <token>', 'Content-Type': 'application/json'},
  body: JSON.stringify({})
};

fetch('https://api.opentype.dev/v1/router/select', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

import requests

url = "https://api.opentype.dev/v1/router/select"

payload = {}
headers = {
    "Authorization": "Bearer <token>",
    "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

{
  "id": "rtr_0f8e3c1a9b2d4e5f8a7b6c5d4e3f2a1b",
  "run_id": "<string>",
  "policy": "<string>",
  "model": {
    "id": "gpt-6-sol",
    "name": "GPT-6 Sol",
    "provider": "openai",
    "open_weights": true
  },
  "classification": {
    "domain": {
      "label": "<string>",
      "probabilities": {}
    },
    "difficulty": {
      "label": "<string>",
      "probabilities": {}
    },
    "task_type": {
      "label": "code_generation",
      "family": "<string>",
      "top": [
        {
          "task_type": "code_generation",
          "family": "coding",
          "probability": 123
        }
      ],
      "fixed": true
    },
    "facets": {
      "difficulty_expected": 123,
      "difficulty": {
        "label": "<string>",
        "probabilities": {}
      },
      "output_length": {
        "label": "<string>",
        "probabilities": {}
      },
      "output_tokens_est": 123,
      "needs_tools": 123,
      "needs_vision": 123,
      "safety_sensitive": 123,
      "language": "<string>",
      "input_tokens_est": 1,
      "target_language": "<string>"
    }
  },
  "ranking": [
    {
      "id": "<string>",
      "name": "<string>",
      "provider": "<string>",
      "open_weights": true,
      "score": 123,
      "blended_price_per_mtok": 123,
      "domain_score": 123,
      "expected_quality": 123,
      "uncertainty": 123,
      "estimated_cost_usd": 123,
      "estimated_latency_ms": 123,
      "latency_estimated": true,
      "strengths": [
        {
          "benchmark": "swe_bench_verified",
          "name": "SWE-bench Verified",
          "norm": 123,
          "weight": 123,
          "contribution": 123,
          "value": 123,
          "rank": 1,
          "gap_to_best": 123
        }
      ],
      "weaknesses": [
        {
          "benchmark": "swe_bench_verified",
          "name": "SWE-bench Verified",
          "norm": 123,
          "weight": 123,
          "contribution": 123,
          "value": 123,
          "rank": 1,
          "gap_to_best": 123
        }
      ],
      "imputed": [
        {
          "benchmark": "<string>",
          "rule": "<string>",
          "from": "<string>"
        }
      ]
    }
  ],
  "score_basis": "domain",
  "filters_applied": [
    {
      "filter": "<string>",
      "removed": 1
    }
  ],
  "low_confidence": true,
  "input_tokens_est": 1,
  "reason": "<string>",
  "decision_model": "neon-1.1",
  "catalog_as_of": "<string>",
  "benchmarks_as_of": "<string>",
  "replayed": true,
  "difficulty_floor": 123,
  "threshold": {
    "q_star": 123,
    "r": 123,
    "tau": 123
  },
  "usage": {
    "input_tokens": 1,
    "output_tokens": 1
  },
  "cost_micros": 1
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

{
  "error": {
    "code": "<string>",
    "message": "<string>",
    "request_id": "<string>",
    "violations": [
      "<string>"
    ]
  }
}

Overview

Runs

Model Router

API keys

Usage and quota

Billing

Health

Select a model

Authorizations

Headers

Body

Response