Models

Models are canonical identities, not an inventory of every provider route. Search results are compact cards; the detail endpoint adds overview metadata and counts. Request profiles, specifications and resources separately when you need full detail.

Use source on the profiles section to load one serving source, for example ?source=githubCopilot. The filter is exact and applies before pagination and total_count; other sections reject it. Omit it to read all profiles. Model scores already support their own source filter, and model prices use provider for the serving source.

The default order starts with the highest Intelligence Index, then uses capability scores, ratings and names for deterministic fallbacks. A developer is the author organization; inference_provider is a serving source. Obtain valid filters from Facets. An official ID is preserved independently of the canonical ID and source aliases.

Model classification labels link to the library with model_type, selecting matching primary or secondary classifications before pagination and counts. Use multimodal for MLLM, vlm for image-only VLM, llm for text LLM, or the type ID in another classification link. unknown selects models without source classification or an applicable first-party fallback; omit the filter or use all for all types. Classification uses reported tasks, capabilities and modalities, so a generic multimodal claim alone does not qualify a model as MLLM. Type filters combine with search and organization, capability, modality and price filters.

For a model ID such as google/gemini-2.5-flash, pass google as the publisher segment and gemini-2.5-flash as the model segment. Encode nested slashes in either segment as %2F. Resolved aliases point to the canonical model; unresolved or ambiguous identities are not guessed. The Hugging Face metadata endpoint returns source-native observations rather than card DTOs.

TensorTrace maintains a reviewed first-party knowledge base for missing facts. Generative Llama families without classification evidence default to LLM; explicit source tasks and non-text modality evidence take priority. Applied supplements carry tensortrace_knowledge with registry/rule revisions, the supplemented field and reference URLs. Raw source documents and identities remain unchanged. The same prepared classification is used by type filters and offline catalogs.

On this page

List and search models

GET /api/v1/models

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
qqueryNostringCase-insensitive search text. Omit it to browse the complete matching catalog.
queryqueryNostringAlternative spelling of q; when both are supplied, query takes precedence.
providerqueryNostringModel publisher namespace. For serving-platform filters, use inference_provider instead.
sourcequeryNostringRegistered data-source ID, such as openRouter. This identifies the collecting source, not necessarily the model developer.
pricequeryNostring Default: "all"; Values: all, free, paidPricing class: all, free or paid. Uses comparable basic token prices; an unknown price is not free.
capabilityqueryNostringComma-separated capability identifiers; all must match.
input_modalityqueryNostringMatch an input modality, such as text, image, audio or video. Obtain current values from /facets/models.
model_typequeryNostring Max. length: 64Match a displayed model classification, including secondary types. Use multimodal for MLLM, vlm for image-only VLM, llm for text LLM, or another model type ID from a classification link. unknown selects models without reported classification evidence. Omit or use all for no type filter. Combines with other discovery filters before pagination and counts; an unrecognized ID returns an empty result.
output_modalitiesqueryNostringComma-separated output modalities; matches models with any listed output modality. all disables this filter.
sortqueryNostring Default: "intelligence"Defaults to Intelligence Index descending; unscored models follow using capability, rating and name as deterministic fallbacks. Price sorts use the card-selected USD Input then Output rates, with missing rates last; cache rates do not participate. Tied basic offerings prefer the model developer. Provider-scoped sorts apply the same policy within that source's quotes.
limitqueryNointeger Default: 48; Minimum: 1; Maximum: 100Maximum number of returned rows. The documented default and range apply; numeric pagination limits are clamped unless stated otherwise.
offsetqueryNointeger Default: 0; Minimum: 0Zero-based number of matching rows to skip. Default 0; negative values are clamped to 0.
identity_kindqueryNostring Values: model, familyRestrict identities to a concrete model or a model family; omit to include both.
developerqueryNostringCanonical developer ID; groups reviewed publisher aliases.
inference_providerqueryNostringRegistered inference source ID; selects current source offerings, including price-only records. Adds source-scoped provider_prices; other model fields retain canonical scope.

Request example

curl 'https://tensortrace.net/api/v1/models?limit=1'

Responses

StatusMeaningBody type
200Incrementally prepared local dataModelPage
400Invalid queryErrorResponse
404Not present in the prepared catalogErrorResponse
503Catalog preparation has not completed yetErrorResponse

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

{
    "data": [
        {
            "id": "google/gemini-2.5-flash",
            "canonical_model_id": "google/gemini-2.5-flash",
            "provider": "google",
            "name": "Google: Gemini 2.5 Flash",
            "official_model_id": "gemini-2.5-flash",
            "developer_id": "google",
            "developer_name": "Google",
            "description": "Illustrative model observation; consult live evidence for current specifications.",
            "context_length": 1048576
        }
    ],
    "total_count": 1,
    "limit": 1,
    "offset": 0
}
FieldTypeRequiredMeaning / constraints
dataarray of CatalogModelYesMax. items: 100
total_countintegerYesMinimum: 0
limitintegerYesMinimum: 1; Maximum: 100
offsetintegerYesMinimum: 0

Read original Hugging Face model metadata

GET /api/v1/huggingface/models

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
limitqueryNointeger Default: 500; Minimum: 1; Maximum: 1000Maximum number of returned rows. The documented default and range apply; numeric pagination limits are clamped unless stated otherwise.
offsetqueryNointeger Default: 0; Minimum: 0Zero-based number of matching rows to skip. Default 0; negative values are clamped to 0.

Request example

curl 'https://tensortrace.net/api/v1/huggingface/models?limit=1'

Responses

StatusMeaningBody type
200Source-backed JSON responsearray of object
400Invalid queryNo declared JSON body
404Unknown model/sourceNo declared JSON body
500Database unavailableNo declared JSON body
503Service temporarily unavailableNo declared JSON body

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

[
    {}
]

Type: array of object.

Prepared model overview; expand explicit detail sections as needed

GET /api/v1/models/{provider}/{model}

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
providerpathYesstringCanonical model publisher namespace; URL-encode this path segment.
modelpathYesstringCanonical model name within the publisher namespace. URL-encode nested slashes in this single segment.

Request example

curl 'https://tensortrace.net/api/v1/models/google/gemini-2.5-flash'

Responses

StatusMeaningBody type
200Source-backed JSON responseModelOverview
400Invalid queryNo declared JSON body
404Unknown model/sourceNo declared JSON body
500Database unavailableNo declared JSON body
503Service temporarily unavailableNo declared JSON body

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

{
    "canonical_model_id": "google/gemini-2.5-flash",
    "model": {
        "id": "google/gemini-2.5-flash",
        "canonical_model_id": "google/gemini-2.5-flash",
        "provider": "google",
        "name": "Google: Gemini 2.5 Flash",
        "official_model_id": "gemini-2.5-flash",
        "developer_id": "google",
        "developer_name": "Google",
        "description": "Illustrative model observation; consult live evidence for current specifications.",
        "context_length": 1048576
    },
    "capabilities": [
        "example"
    ],
    "benchmark_score_counts": {},
    "sections": {}
}
FieldTypeRequiredMeaning / constraints
canonical_model_idstringYes
modelCatalogModelYes
capabilitiesarray of stringYes
benchmark_score_countsobjectYes
sectionsobjectYes
referencesarray of ContentReferenceNoExactly anchored model/distribution metadata. Base-model or derived-from relationships do not establish identity. Max. items: 20

Complete prepared profiles, specifications or external resources; default 20 per page

GET /api/v1/models/{provider}/{model}/sections/{section}

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
providerpathYesstringCanonical model publisher namespace; URL-encode this path segment.
modelpathYesstringCanonical model name within the publisher namespace. URL-encode nested slashes in this single segment.
sectionpathYesstring Values: profiles, specifications, resourcesPrepared model section: profiles (platform/generation metadata), specifications (source assertions) or resources (external links).
sourcequeryNostringExact serving source ID, such as githubCopilot or azureFoundry. Applies to profiles only; other sections reject this filter. Pagination and total_count apply after filtering. Omit to read all platform profiles.
limitqueryNointeger Default: 20; Minimum: 1; Maximum: 100Maximum number of returned rows. The documented default and range apply; numeric pagination limits are clamped unless stated otherwise.
offsetqueryNointeger Default: 0; Minimum: 0Zero-based number of matching rows to skip. Default 0; negative values are clamped to 0.

Request example

curl 'https://tensortrace.net/api/v1/models/google/gemini-2.5-flash/sections/profiles?limit=1'

Responses

StatusMeaningBody type
200Current local observationsDetailSectionPage
400Invalid requestNo declared JSON body
404Unknown model or sectionNo declared JSON body

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

{
    "data": [
        {}
    ],
    "total_count": 1,
    "limit": 1,
    "offset": 0
}
FieldTypeRequiredMeaning / constraints
dataarray of objectYesMax. items: 100
total_countintegerYesMinimum: 0
limitintegerYesMinimum: 1; Maximum: 100
offsetintegerYesMinimum: 0

Read a dated offline client catalog manifest

GET /api/v1/sync/manifest

Separate client distribution; ordinary API reads remain current. Compare SHA-256 objects and replace changed buckets. The serial worker publishes at most every 15 minutes. Includes all model cards and organizations plus top 100 entries per core ranking; large llmPerf, Kaggle and historical archive observations remain on-demand. Objects publish before the manifest atomically.

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
If-None-MatchheaderNostringETag from the last successfully committed local catalog; permits an empty 304 response.

Request example

curl 'https://tensortrace.net/api/v1/sync/manifest'

Responses

StatusMeaningBody type
200Offline schema 1 manifest. Hashes cover compressed gzip bytes.OfflineManifest
304Unchanged manifest; empty body.No declared JSON body
503No client distribution published yet.No declared JSON body

Response example · 200

Illustrative values; schema-valid and not a live catalog measurement.

{
    "schema": 1,
    "generation": "example",
    "generated_at": "2026-10-05T00:00:00Z",
    "model_count": 1,
    "chunks": [
        {
            "id": "google/gemini-2.5-flash",
            "kind": "models",
            "sha256": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
            "bytes": 1,
            "rows": 1
        }
    ]
}
FieldTypeRequiredMeaning / constraints
schemaintegerYes
generationstringYes
generated_atstring · date-timeYes
model_countintegerYes
chunksarray of objectYes

Download an immutable compressed catalog bucket

GET /api/v1/sync/objects/{hash}

Access: public read; no TensorTrace API key.

Parameters

NameLocationRequiredType / limitsMeaning
hashpathYesstring Pattern: ^[0-9a-f]{64}$Lowercase SHA-256 from the manifest, covering the compressed object bytes.

Request example

curl 'https://tensortrace.net/api/v1/sync/objects/EXAMPLE_ID'

Responses

StatusMeaningBody type
200SHA-256 addressed gzip JSON array. Models: id, model and selectors (search, types, modalities, membership, price and rank), offers. Boards: normalized ranking fields and entries. Metadata: directories and model facets. Validate byte count, hash and manifest row count before a local transaction. Cache-Control: immutable, one year.string · binary
404Object not found.No declared JSON body

Response example · 200

The response contains original bytes or an image. Content types: application/gzip. Preserve the declared encoding and checksum; do not parse it as JSON.

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

Install TensorTrace