OpenRouter API Analysis - Current Model Structure

December 16, 2025 · View on GitHub

Endpoint Information

URL: https://openrouter.ai/api/v1/models Method: GET Authentication: Bearer token in Authorization header

Query Parameters (Currently Supported)

category - Filter models by category
supported_parameters - Filter by supported parameters
use_rss - Return as RSS feed
use_rss_chat_links - Include chat links in RSS

Response Structure

Root Object

{
  "data": [/* array of model objects */]
}

Model Object Fields

Basic Identifiers

Field	Type	Description	Example
`id`	string	Full model identifier with variant	"nvidia/nemotron-3-nano-30b-a3b:free"
`canonical_slug`	string	Base identifier without variant	"nvidia/nemotron-3-nano-30b-a3b"
`name`	string	Human-readable display name	"NVIDIA: Nemotron 3 Nano 30B A3B (free)"
`hugging_face_id`	string	Optional HuggingFace model ID	"nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16"
`created`	integer	Unix timestamp of model addition	1765731275

Core Properties

Field	Type	Description	Example
`description`	string	Detailed model description and notes	"NVIDIA Nemotron 3 Nano..."
`context_length`	integer	Maximum context window in tokens	256000

Architecture Object

Field	Type	Description	Example
`modality`	string	Input/output format	"text->text", "text+image->text"
`input_modalities`	array[string]	Supported input types	["text", "image", "file"]
`output_modalities`	array[string]	Supported output types	["text", "image", "embeddings"]
`tokenizer`	string	Tokenizer type	"GPT", "Other"
`instruct_type`	string\|null	Instruction format	null

Pricing Object

All prices are per token unless noted.

Field	Type	Description	Example
`prompt`	string	Input token price	"0.00000175"
`completion`	string	Output token price	"0.000014"
`request`	string	Per-request price	"0"
`image`	string	Per-image price	"0"
`web_search`	string	Web search feature price	"0.01"
`internal_reasoning`	string	Reasoning token price	"0"
`input_cache_read`	string	Cache read price	"0.000000175"

Notes:

Free models have "0" for all pricing fields
Prices are in USD
To calculate per-1K tokens: multiply by 1000

Top Provider Object

Field	Type	Description
`context_length`	integer	Provider's actual context limit
`max_completion_tokens`	integer\|null	Maximum output tokens per request
`is_moderated`	boolean	Whether content moderation is enabled

Supported Parameters

Array of strings indicating which API parameters the model supports:

Common Parameters:

temperature - Randomness control
top_p - Nucleus sampling
top_k - Top-K sampling
frequency_penalty - Repetition penalty
presence_penalty - Token presence penalty
repetition_penalty - Alternative repetition control
min_p - Minimum probability threshold
top_a - Top-A sampling

Advanced Features:

tools - Function/tool calling
tool_choice - Control which tool to use
response_format - Structured output format
structured_outputs - JSON schema enforcement
reasoning - Chain-of-thought reasoning
include_reasoning - Include reasoning in response
seed - Deterministic generation
max_tokens - Output length limit
stop - Stop sequences
logit_bias - Token probability biasing
logprobs - Return token probabilities
top_logprobs - Number of top logprobs to return

Specialized:

prompt_truncate_len - Prompt truncation
transforms - Input transformations

Default Parameters Object

Field	Type	Description
`temperature`	float\|null	Default temperature
`top_p`	float\|null	Default top_p
`frequency_penalty`	float\|null	Default frequency penalty

Per Request Limits

Can be null or contain:

Maximum prompt tokens per request
Maximum completion tokens per request

Sample Model Records

Free Model Example

{
  "id": "nvidia/nemotron-3-nano-30b-a3b:free",
  "name": "NVIDIA: Nemotron 3 Nano 30B A3B (free)",
  "context_length": 256000,
  "pricing": {
    "prompt": "0",
    "completion": "0",
    "request": "0"
  },
  "supported_parameters": [
    "include_reasoning",
    "max_tokens",
    "reasoning",
    "seed",
    "temperature",
    "tool_choice",
    "tools",
    "top_p"
  ]
}

Paid Model Example

{
  "id": "openai/gpt-5.2-chat",
  "name": "OpenAI: GPT-5.2 Chat",
  "context_length": 128000,
  "pricing": {
    "prompt": "0.00000175",
    "completion": "0.000014",
    "web_search": "0.01",
    "input_cache_read": "0.000000175"
  },
  "architecture": {
    "modality": "text+image->text",
    "input_modalities": ["file", "image", "text"],
    "output_modalities": ["text"]
  },
  "top_provider": {
    "max_completion_tokens": 16384,
    "is_moderated": true
  }
}

Searchable/Filterable Fields

Based on the structure, we can filter/search by:

Direct Filters

Provider: Extract from id (e.g., "openai/", "anthropic/", "nvidia/")
Free models: pricing.prompt == "0" && pricing.completion == "0"
Context length: context_length >= X
Moderation: top_provider.is_moderated
Creation date: created >= timestamp

Capability Filters

Multimodal: Check architecture.input_modalities length > 1
Vision: "image" in architecture.input_modalities
File input: "file" in architecture.input_modalities
Image output: "image" in architecture.output_modalities
Tool calling: "tools" in supported_parameters
Reasoning: "reasoning" in supported_parameters
Structured output: "structured_outputs" in supported_parameters

Price-Based Filters

Prompt cost: Parse pricing.prompt as float
Completion cost: Parse pricing.completion as float
Total cost: Combined prompt + completion
Has web search: pricing.web_search != "0"
Has caching: pricing.input_cache_read != "0"

Text Search

Name: Full-text search in name field
Description: Search in description field
Provider: Search in id field
Model family: Search for "gpt", "claude", "llama", etc.

Useful Derived Metrics

Cost Per Million Tokens

prompt_per_1M = parseFloat(pricing.prompt) * 1_000_000
completion_per_1M = parseFloat(pricing.completion) * 1_000_000
total_per_1M = prompt_per_1M + completion_per_1M

Capability Score

Count of supported parameters = indicator of feature richness

Efficiency Metrics

Tokens per dollar: context_length / (prompt_per_1M + completion_per_1M)
Relative cost: Compare to baseline (e.g., GPT-4 Turbo)

Categorization

Based on name/description/capabilities:

"Instruct/Instructional" models
"Chat" models
"Vision" models
"Fast/Turbo" models
"Preview/Experimental" models
"Free" models

Common Query Patterns

Find Cheapest Text Models

Filter: modality == "text->text"
Sort: prompt + completion price ascending

Find Free Multimodal Models

Filter: pricing.prompt == "0" AND input_modalities.length > 1
Sort: context_length descending

Find Models With Reasoning

Filter: "reasoning" in supported_parameters
Sort: created descending (newest first)

Find Instructional Models

Filter: name.includes("Instruct") OR description.includes("instruction")
Sort: pricing ascending

Find High-Context Models

Filter: context_length >= 100000
Sort: context_length descending

Notes for Implementation

Pricing as strings: All pricing values are strings, need parseFloat() for comparisons
Null handling: Many fields can be null (hugging_face_id, instruct_type, max_completion_tokens)
Provider extraction: Parse id field (split on first "/")
Free model detection: Check if prompt AND completion are "0"
Capability detection: Search arrays (input_modalities, output_modalities, supported_parameters)
Date handling: created is Unix timestamp, convert for display
Modality parsing: Format is "input->output" or "input+input->output"