- activebooleanrequired
- boundbooleanrequired
- Whether the caller can call this model right now: a global active
binding, or a binding in the caller's own tenant. Unlike `active`
(any tenant's binding counts) and `deployability.status == callable`
(same), this is answered for the caller who asked.
- capabilitiesstring[]required
- context_windowinteger · int32
- created_atstring · date-timerequired
- deployabilityDeployabilityrequired
- How (and whether) this model can be served — see [`Deployability`].
- blueprintsstring[]required
- Names of the Blueprints that can launch this model's family.
- est_hourly_centsinteger · int64
- Floor hourly cost of a self-hosted launch, integer cents. Present
only when a Blueprint AND a priced GPU both exist.
- gpu_requiredbooleanrequired
- Whether serving this model needs a GPU launch (i.e. it has a
Blueprint) versus being callable through an existing binding.
- statusDeployabilityStatusrequired
- Deployability tiers, in descending readiness. The wire strings are the
`?deployability=` filter values.
- display_namestringrequired
- familystringrequired
- hf_repostring
- idstringrequired
- kindstring
- The binding kind — which inference surface serves it (`openai_chat`,
`openai_embed`, `openai_transcription`, `openai_image`, …); `null`
when `bound` is false.
- licensestring
- max_output_tokensinteger · int32
- The most tokens one reply may carry, as the provider states it.
`None` ⇒ unpublished; a consumer must not guess it.
- modalitystringrequired
- params_total_bnumber · double
- pricingPricingrequired
- The list price, every dimension the cost writer bills on. The three cache
rates are published so a consumer that mirrors this catalog and estimates
its own spend (Lighthouse does) prices a cache read and a cache write with
the SAME numbers the ledger uses, instead of guessing a multiplier and
drifting from the bill. Absent when the model has no such rate.
- billing_unitBillingUnit | null
- The unit the ledger charges this model in, for the binding the caller
resolves to — derived by `model_registry::billing_unit`, the same
decision the cost writers take. Null when the caller has no binding
for this model (nothing it could be charged for).
- cache_creation_1hr_cents_per_mtoknumber · double
- cache_creation_cents_per_mtoknumber · double
- cached_input_cents_per_mtoknumber · double
- cents_per_audio_minutenumber · double
- Per minute of audio (speech generation and transcription). Null when unset.
- cents_per_classificationnumber · double
- Per input classified. Null when unset.
- cents_per_image_baselinenumber · double
- Per generated image at baseline quality. Null when unset.
- cents_per_mcharnumber · double
- Per million input characters synthesized (character-priced speech).
Null when unset.
- cents_per_requestnumber · double
- Per request (moderation, fixed-cost custom endpoints). Null when unset.
- cents_per_search_unitnumber · double
- Per rerank search unit. Null when unset.
- cents_per_video_secondnumber · double
- Per generated second of video. Null when unset.
- currencystringrequired
- image_input_cents_per_mtoknumber · double
- Rate for input tokens that are images (gpt-image edits' reference
images). Absent means image input bills at `input_cents_per_mtok`.
- input_cents_per_mtoknumber · double
- long_contextLongContextRates | null
- The provider's long-prompt tier: once a prompt exceeds
`above_prompt_tokens` the whole call bills at these rates. Absent when
the model has one rate at every length.
- output_cents_per_mtoknumber · double
- tiersTierRatesrequired
- The rate tiers a call to this model can be billed at through Nozzle,
in permille of the standard rate: `flex_permille` / `priority_permille`
(ask with `service_tier`), `batch_permille` (lines of a `/v1/batches`
batch), and an `off_peak` clock (`{permille, peak_utc}`: every hour
outside the peak windows is billed at permille). Null means the tier
is not available for this model.
- batch_permilleinteger · uint16
- flex_permilleinteger · uint16
- off_peakOffPeak | null
- priority_permilleinteger · uint16
- providerstring
- The upstream provider the caller's resolving binding routes to;
`null` when `bound` is false.
- reasoning_effort_maparray
- The same efforts, each with the request fragment it becomes upstream
(`{"reasoning_effort": "low"}`, `{"output_config": {"effort": "low"}}`,
`{"thinkingConfig": {"thinkingBudget": 0}}`) and the operator's note
where the word alone would mislead.
- reasoning_effortsarray
- The `reasoning_effort` values the router accepts for the caller's
resolving binding, least thinking first, read from the same row the
router validates against. `[]` when the model takes none (a value is
refused by name). `null` when `bound` is false, or when Nozzle
forwards the word verbatim and the upstream decides, so there is no
list Nozzle can vouch for.
- reasoning_visibleReasoningVisibility | null
- What readable reasoning a stream carries as `delta.reasoning_content`:
`summary` (a provider-written summary, which Nozzle requests),
`full` (the provider returns the reasoning text itself) or `none`.
Never encrypted or redacted reasoning. `null` when `bound` is false.
- updated_atstring · date-timerequired
- upstream_modelstring
- The model id on that provider's wire; `null` when `bound` is false.
- wire_capabilitiesWireCapabilities | null
- The parameter support the router enforces for the caller's resolving
binding — the same booleans `GET /v1/models/{id}/capabilities`
answers, computed the same way (wire family, narrowed by the provider
row, the model's own profile and the binding kind). `null` when
`bound` is false: there is no binding to answer for. Distinct from
`capabilities`, which is the curated tag list a discovered model
carries before any binding exists.