Catalog, benchmarks and capability
Mirrored model metadata, Hugging Face weight licences, route speed measurements, multi-seller endpoint offers, openly licensed benchmark boards, and Vitna's IRT capability scale. Kept separate from /api/v1/markets so meta.provenance always distinguishes mirrored third-party data from Witness first-party provider observations.
List catalog models
GET /api/v1/catalog/modelsOptional query parameters: author, modality, capability (tools, reasoning, structured), open_weights (true or false), q, and limit (1..2000).
{
"models": [
{
"id": "deepseek/deepseek-chat",
"urlSlug": "deepseek--deepseek-chat",
"author": "deepseek",
"authorName": "DeepSeek",
"slug": "deepseek-chat",
"name": "DeepSeek V3",
"contextLength": 163840,
"inputModalities": [ ... ],
"outputModalities": [ ... ],
"pricing": {
"input": 0.3,
"output": 0.88
},
"huggingFaceId": "deepseek-ai/DeepSeek-V3",
"weights": {
"repo": "deepseek-ai/DeepSeek-V3",
"license": "mit"
},
"sellerCount": 5
}
],
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "mirror",
"count": 340,
"totalCatalog": 340,
"sources": [
{
"id": "openrouter-models",
"url": "https://openrouter.ai/api/v1/models",
"retrievedAt": "2026-09-12T00:00:00.000Z",
"attribution": "OpenRouter public model list"
}
]
}
}Single catalog model
GET /api/v1/catalog/models/{slug}Accepts {slug} as author--slug (deepseek--deepseek-chat), URL-encoded author%2Fslug, or bare slug.
{
"model": {
"id": "deepseek/deepseek-chat",
"urlSlug": "deepseek--deepseek-chat",
"name": "DeepSeek V3",
"contextLength": 163840,
"pricing": {
"input": 0.3,
"output": 0.88
},
"speedEndpoints": [ ... ],
"sellers": [ ... ]
},
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "mirror",
"sources": [ ... ]
}
}Model seller offers
GET /api/v1/catalog/models/{slug}/sellersReturns every seller endpoint mirrored for one model, with list input/output rates, declared quantization, context window, and dispersion summary.
{
"modelId": "deepseek/deepseek-chat",
"urlSlug": "deepseek--deepseek-chat",
"listedPricing": {
"input": 0.3,
"output": 0.88
},
"sellers": [
{
"provider": "DeepInfra",
"input": 0.28,
"output": 0.88,
"quantization": "fp8",
"contextLength": 163840
}
],
"summary": {
"sellerCount": 5,
"cheapestInputUsd": 0.25,
"dearestInputUsd": 0.8,
"spreadRatio": 3.2,
"quantizations": [ ... ]
},
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "mirror",
"notice": "Quantization is the seller's own declaration.",
"sources": [ ... ]
}
}Witness SKU capability profile
GET /api/v1/catalog/skus/{slug}/capabilitiesResolves a Witness SKU (provider/slug or bare slug, such as claude-sonnet-4-6 or gpt-5.4) into its three-provenance capability profile: declared_mirror capabilities and context window, measured_third_party throughput and TTFT with sample size and retrieval timestamp, curated Witness providerTerms when present, and the engine's admissibility contract (admissibility). See Three-provenance capability eligibility.
{
"sku": {
"id": "anthropic/claude-sonnet-4-6",
"slug": "claude-sonnet-4-6",
"name": "Claude Sonnet 4.6",
"classId": "model",
"providerId": "anthropic",
"catalogListingId": "anthropic/claude-sonnet-4.6"
},
"provenanceClasses": {
"declared_mirror": {
"provenance": "declared_mirror",
"listingId": "anthropic/claude-sonnet-4.6",
"capabilities": [ ... ],
"contextLength": 200000,
"source": "https://openrouter.ai/api/v1/models",
"retrievedAt": "2026-09-12T00:00:00.000Z"
},
"measured_third_party": {
"provenance": "measured_third_party",
"listingId": "anthropic/claude-sonnet-4.6",
"measurer": "OpenRouter speed mirror",
"throughputToksSec": 46.7,
"latencyTtftMs": 786,
"requests": 58622,
"windowMinutes": 30,
"retrievedAt": "2026-09-12T00:00:00.000Z"
},
"measured_customer": {
"provenance": "measured_customer",
"metricKey": "verifiedCapabilities",
"requiredFields": [ ... ]
}
},
"providerTerms": { ... },
"admissibility": {
"declared_mirror": {
"provenance": "declared_mirror",
"answersConstraints": [ ... ],
"cannotAnswerConstraints": [ ... ]
},
"measured_third_party": {
"provenance": "measured_third_party",
"answersConstraints": [ ... ],
"cannotAnswerConstraints": [ ... ]
},
"measured_customer": {
"provenance": "measured_customer",
"answersConstraints": [ ... ],
"cannotAnswerConstraints": [ ... ]
}
},
"knownCapabilityTokens": [ ... ],
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "three_class_profile",
"sources": [ ... ]
}
}Openly licensed benchmark boards
GET /api/v1/catalog/benchmarksOptional query parameters: class, board, and summary=true. Republishes only boards whose results licence was verified by platform/boards/sources.mjs, carrying each publisher's licence, runBy, metric, and mandatory notice.
{
"boards": [
{
"id": "aider",
"label": "Aider polyglot",
"classId": "model",
"licence": "Apache-2.0",
"runBy": "publisher",
"metric": "pass rate (%)",
"notice": "Source: Aider LLM Leaderboards",
"url": "https://aider.chat/docs/leaderboards/",
"retrievedAt": "2026-09-12T00:00:00.000Z",
"columns": [ ... ],
"modelCount": 60,
"rowCount": 69,
"rows": [ ... ]
}
],
"unread": [ ... ],
"unreadable": [ ... ],
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "mirror",
"count": 6,
"sources": [ ... ]
}
}Fitted capability scale
GET /api/v1/catalog/capabilityOptional query parameters: ranked_only=true and board. Serves Vitna's IRT capability scale fit over the openly licensed benchmark mirrors (mean = 100, standardDeviation = 15), together with column difficulties, board ties, and excluded disconnected groups.
{
"runs": [
{
"id": "openai/gpt-5.4",
"index": 118.4,
"columns": 22,
"boards": 4,
"ranked": true
}
],
"columns": [
{
"id": "aider:polyglot",
"difficulty": 98.2,
"discrimination": 1.4
}
],
"ties": [ ... ],
"excluded": [ ... ],
"scale": {
"mean": 100,
"standardDeviation": 15,
"minColumns": 3,
"minShared": 3
},
"meta": {
"generatedAt": "2026-11-02T00:00:00.000Z",
"source": "snapshot",
"provenance": "derived_over_mirror",
"runCount": 120,
"rankedCount": 85,
"columnCount": 40,
"sources": [ ... ]
}
}