Prices API
HIPS publishes its prices for machines to read. Three routes on https://hips.com answer the public catalog with every published rate, one model with its own rates, and the history of each rate: the same reads the Pricing page and each model's page are built from. Three price feeds on https://ai.hips.com list each pool's offers, one document per model, under the model id that runs it there (The price feeds per pool). Use them to show HIPS's prices in your own product, to estimate what a workload costs, or to follow a rate as it moves.
| Route on https://hips.com | Answers | Key |
|---|---|---|
GET /api/models | The whole public catalog: every published model, every public price row in effect or announced, the plans and the public settings | None |
GET /api/models/{id} | One published model and its own price rows | None |
GET /api/prices/history | How each published on-demand rate has moved, per model, pool and kind | None |
- Route on https://hips.com
GET /api/models- Answers
- The whole public catalog: every published model, every public price row in effect or announced, the plans and the public settings
- Key
- None
- Route on https://hips.com
GET /api/models/{id}- Answers
- One published model and its own price rows
- Key
- None
- Route on https://hips.com
GET /api/prices/history- Answers
- How each published on-demand rate has moved, per model, pool and kind
- Key
- None
Each answers JSON in a data envelope, the same for every caller: no key, no cookie and no sign-in, and nothing about a workspace. The answers carry no cross-origin headers, so read them from your server, not from a browser page on another site. They are the catalog's rates, the same for every workspace; what one request was charged is in its own answer's x-hips-cost-micros (How a request is priced).
Every amount is a whole number, amount_micros, in micro-dollars: millionths of a US dollar, 1,000,000 to the dollar. A row's unit says what one amount buys, and its kind what it prices.
| Kind | Unit | What one amount buys |
|---|---|---|
language_input | per_million_tokens | A million prompt tokens of a language model |
language_input_cached | per_million_tokens | A million prompt tokens billed as verified cache hits. HIPS keeps no prompt cache yet, so cached input is billed as input today: every prompt token is priced at language_input (Prompt caching) |
language_output | per_million_tokens | A million generated tokens |
classification | per_thousand_classifications | A thousand classifications |
task_second | per_second | A started second of a task model's compute |
media_unit | per_media_unit | One unit of the setting the row names in setting |
always_on_hourly | per_allocated_hour | An hour of always-on capacity allocated |
allocation_storage | per_gigabyte_hour | A gigabyte kept for an hour for a stopped allocation |
storage | per_gigabyte_month | A gigabyte stored for a month (Storage) |
plan_fee | per_month | A paid plan's monthly fee |
plan_fee_annual | per_year | A paid plan's yearly fee |
plan_included_credits | per_period | The credit a paid plan includes each monthly period, granted as a weekly pool (Weekly pool of included credit) |
batch_discount | basis_points | Not money: the batch discount in hundredths of a percent, 5,000 being 50% |
mto_discount | basis_points | Not money: a payment discount in hundredths of a percent |
- Kind
language_input- Unit
per_million_tokens- What one amount buys
- A million prompt tokens of a language model
- Kind
language_input_cached- Unit
per_million_tokens- What one amount buys
- A million prompt tokens billed as verified cache hits. HIPS keeps no prompt cache yet, so cached input is billed as input today: every prompt token is priced at
language_input(Prompt caching)
- Kind
language_output- Unit
per_million_tokens- What one amount buys
- A million generated tokens
- Kind
classification- Unit
per_thousand_classifications- What one amount buys
- A thousand classifications
- Kind
task_second- Unit
per_second- What one amount buys
- A started second of a task model's compute
- Kind
media_unit- Unit
per_media_unit- What one amount buys
- One unit of the setting the row names in
setting
- Kind
always_on_hourly- Unit
per_allocated_hour- What one amount buys
- An hour of always-on capacity allocated
- Kind
allocation_storage- Unit
per_gigabyte_hour- What one amount buys
- A gigabyte kept for an hour for a stopped allocation
- Kind
storage- Unit
per_gigabyte_month- What one amount buys
- A gigabyte stored for a month (Storage)
- Kind
plan_fee- Unit
per_month- What one amount buys
- A paid plan's monthly fee
- Kind
plan_fee_annual- Unit
per_year- What one amount buys
- A paid plan's yearly fee
- Kind
plan_included_credits- Unit
per_period- What one amount buys
- The credit a paid plan includes each monthly period, granted as a weekly pool (Weekly pool of included credit)
- Kind
batch_discount- Unit
basis_points- What one amount buys
- Not money: the batch discount in hundredths of a percent, 5,000 being 50%
- Kind
mto_discount- Unit
basis_points- What one amount buys
- Not money: a payment discount in hundredths of a percent
A kind appears only where a row of it is published. Markups, commissions, provider payouts and floors are never in these answers.
A language_input row with amount_micros 137200 and unit per_million_tokens:
- Per million tokens: 137,200 / 1,000,000 = $0.1372.
- Per token: 137,200 / 1,000,000 / 1,000,000 = $0.0000001372, that is
amount_microsdivided by 1,000,000,000,000. - For a prompt of 2,000 tokens: 2,000 x 137,200 / 1,000,000 = 274.4 micro-dollars, $0.0002744, at that rate.
- A
storagerow of 100000per_gigabyte_monthis $0.10 per GB-month; aplan_feerow of 20000000per_monthis $20 a month.
A request runs in one of two pools, and each pool carries its own price rows for a model, named in these answers as on the inference API. A request chooses among three:
| Choice | Value | Where it runs |
|---|---|---|
| Auto (default) | auto | Wherever HIPS has capacity for the model now, in the Certified or the Community pool, at that pool's rates; Auto has no price rows of its own |
| Certified | certified | Data-center supply operated by an approved provider |
| Community | community | Devices their owners run |
- Choice
- Auto (default)
- Value
auto- Where it runs
- Wherever HIPS has capacity for the model now, in the Certified or the Community pool, at that pool's rates; Auto has no price rows of its own
- Choice
- Certified
- Value
certified- Where it runs
- Data-center supply operated by an approved provider
- Choice
- Community
- Value
community- Where it runs
- Devices their owners run
- A request that names no pool takes its key's default pool, and Auto where the key sets none (Requirements and headers).
- Under Auto, HIPS chooses among the pools the workspace and its guardrails allow that offer and price the model and have capacity for it now. Where both have, it weighs the request's expected charge in each pool against its expected time to the first token, the lower price winning a tie and then the Certified pool; where the first token cannot be estimated in a pool, the lower price decides. A workspace that keeps its requests out of the Community pool runs Auto in the Certified pool alone.
- The request pays the rates of the pool it ran in, and the answer's
x-hips-poolnames that pool. - A model's price row whose
poolis null applies in every pool the model is offered in.
On chat completions, Responses and Messages, a model id may carry its pool. A model id without a suffix uses Auto, as in the Playground and everywhere else in HIPS, or the key's default pool where the workspace set one.
| Model id | Runs |
|---|---|
<model id> | Under the key's default pool: Auto, unless the workspace set another |
<model id>:auto | Auto, whatever the key's default |
<model id>:certified | The Certified pool alone, with no fallback, at its rates |
<model id>:community | The Community pool alone, with no fallback, at its rates |
- Model id
<model id>- Runs
- Under the key's default pool: Auto, unless the workspace set another
- Model id
<model id>:auto- Runs
- Auto, whatever the key's default
- Model id
<model id>:certified- Runs
- The Certified pool alone, with no fallback, at its rates
- Model id
<model id>:community- Runs
- The Community pool alone, with no fallback, at its rates
A bare id, or :auto, is served wherever HIPS has capacity for the model, in the Certified or the Community pool, among the pools your workspace's settings allow. Where both have capacity, it takes the one where the request costs less, weighing the expected time to the first token where both pools can estimate it.
- A suffix is the request that names the model without it, with
x-hips-poolset toauto,certifiedorcommunity, and for a poolx-hips-fallback: false(Requirements and headers).x-hips-poolnaming anything else beside it, orx-hips-fallback: truebeside:certifiedor:community, is a 400requirement_conflict. A workspace that keeps its requests out of the Community pool refuses:communitywith the same code atmodel. x-hips-modeland the usage record name the model without the suffix;x-hips-poolnames the pool the request ran in.- Only these three words are pool suffixes. Any other, or one on a router's name or a
dep_id, is read as the whole name: a 404model_not_found, or the router's own 400. The task API, embeddings, rerank and a batch file's lines take no suffix: name the pool withx-hips-poolorhips.poolthere.
revisioncounts the catalog's changes: it moves whenever HIPS changes a model, a pool, a price, a plan or a setting.prices_updated_atis the latest moment a price row now in effect took effect. An announced price moves it when its moment comes, with no new revision; announcing one does not move it. It is null while no price is in effect.published_atis when the catalog was last published. It moves on a regular cadence even when nothing changed, so it never tells you that anything did.- The site keeps its copy of the catalog for 60 seconds before reading it again, and answers
Cache-Control: public, max-age=60; the price history answers what is left of its own 60 seconds. Reading more often than once a minute gains nothing: comparerevisionandprices_updated_atto tell whether anything changed. - There is no
ETag: every request is answered in full. Every answer the catalog served carriesx-hips-request-id; quote it when you write to support.
https://hips.com/api/modelsservedThe whole public catalog in one answer: every published model with its pools, every public price row in effect or announced ahead, the plans a new subscriber may choose and the public settings.
cURL
curl -s https://hips.com/api/modelsParts list: The answer's data
revisionintegerThe catalog's revision (How often it changes).
published_atRFC 3339 timeWhen the catalog was last published.
prices_updated_atRFC 3339 time or nullThe latest moment a price row now in effect took effect.
modelsarrayEvery model in
preview,liveordeprecated; a draft or a retired model is never listed.pricesarrayEvery public price row in effect now and every one announced ahead, on demand and batch.
plansarrayThe plans a new subscriber may choose, with their terms.
stable_schedulesarrayAn empty list: no stable price is published in the catalog (Stable prices).
settingsobjectThe public settings (Settings).
Parts list: Each row of prices
kindstringWhat the row prices (Money and units).
modelstring or nullThe model's id, as the model's
idand as a request'smodel; null on a row that is not a model's (a plan's, storage, a discount).poolstring or nullcertifiedorcommunity(Pools); null on a model's row that applies in every pool it is offered in, and on every row that is not a model's.settingstring or nullOn a
media_unitrow, the model's setting the rate applies to; null on every other row.planstring or nullOn
plan_fee,plan_fee_annualandplan_included_credits, the plan the row prices (pro,business); null on every other row.unitstringWhat one amount buys (Money and units).
amount_microsintegerThe amount in micro-dollars, or in basis points where
unitisbasis_points.batchbooleanTrue on a batch row: the rate a batch line pays, derived from the on-demand row of the same model, pool, kind and setting at the batch discount, 50%, rounded down to the micro-dollar, and published beside it at the same
effective_from.pricing_modestringmarketon every row of the catalog.effective_fromRFC 3339 timeWhen the row took effect, or takes effect when
scheduledis true.scheduledbooleanTrue while
effective_fromis ahead: a change announced before it applies.
- For one model, pool, kind and setting, the row with
batchandscheduledboth false is the on-demand rate in effect now, and the row withbatchtrue andscheduledfalse beside it is the batch rate. - A row with
scheduledtrue is an announced change. The row in effect stays listed beside it until itseffective_from; from that moment the new row is in effect and the one it replaced is no longer listed. - A request's rate is the one in effect when it is accepted, and holds for every attempt and retry of it.
- A model's batch rows are listed whether or not it is sold in batch: where its
batch.offeredis false, a batch for it is refused with 400batch_not_offered. - A
language_input_cachedrow is the rate a verified cache hit will be billed at. A pool bills it only where the model'sprompt_cacheflag for that pool is true; both flags read false today, since HIPS keeps no prompt cache yet, so every prompt token is billed atlanguage_input.
Reading one rate, in dollars per unit:
import jsonimport osimport urllib.request with urllib.request.urlopen("https://hips.com/api/models") as answer: catalog = json.load(answer)["data"] def rate(model, pool, kind): """The on-demand rate in effect now, in dollars per unit, or None.""" rows = [p for p in catalog["prices"] if p["model"] == model and p["kind"] == kind and not p["batch"] and not p["scheduled"]] row = next((p for p in rows if p["pool"] == pool), None) or next((p for p in rows if p["pool"] is None), None) return None if row is None else row["amount_micros"] / 1_000_000 # Dollars per million input tokens in the Certified pool.print(rate(os.environ["HIPS_MODEL"], "certified", "language_input"))Parts list: On each model
idstringThe model id a request names, and the
modelof its price rows.statuspreview,liveordeprecatedA deprecated model carries the day it retires in
retires_at.typestringThe model's type; a
languagemodel is priced per token.billingobjectunitandkinds: the kinds the model is billed in, which its price rows carry. A language model's arelanguage_input,language_input_cachedandlanguage_output.poolsarrayEach pool row that serves the model now:
pool,lane(shared, oralways_on),engine(for examplellamacpporvllm) andprecision, the precision that pool serves the model at on that engine:nameas recorded (Q4_K_M,mxfp4,bf16) andlevel, one offp32,bf16,fp16,fp8,int8,fp6,fp4andint4, or null for a type with no such word.precisionis left out where none is recorded. A pool's rates are the same on each of its engines. A model can carry price rows for a pool itspoolsdoes not list: they are that pool's rates once it serves the model, and no request runs there meanwhile.prices_publishedbooleanTrue when at least one pool has a row in effect for every kind
billing.kindsnames. A model listed with false has no complete price yet.batchobjectoffered, whether a batch of the model is accepted, andreason_code,floorsorotherwhere batch was turned off for it, else null.prompt_cacheobjectPer pool,
certifiedandcommunity: true where a cached prompt token is billed atlanguage_input_cached. False in both pools today: cached input is billed as input.limitsobjectThe limits that are set, by name,
context_tokensamong them.
Each model also carries its names and description (name, family, size, variant, description), its task and modality, its dates (released_at, deprecated_at), its source on Hugging Face (hugging_face_id, hugging_face_revision, hugging_face_url), architecture, parameter_count, the current version and pinned_id, its versions and aliases, modalities, formats, license, tool_calls per pool, and agents, the coding agents tested with it (Models and task guides). None of them changes a price.
Parts list: Each plan
planstringfree,proorbusiness.termsobject or nullThe terms in effect for a new subscriber:
models, the model ids the plan includes (null: every model),features, amongrouter,guardrails,knowledgeanddeployments, andlimitsby name, empty while none is set.effective_fromRFC 3339 time or nullWhen those terms took effect.
scheduledarrayEach change of terms announced ahead, with its
effective_fromandterms.
A plan's fees and its included credit are price rows naming it in plan: plan_fee per month, plan_fee_annual per year and plan_included_credits per monthly period. Free has no fee row. Billing explains the plans.
settings holds the catalog's public settings, each only once it is set: the free allowance (free_allowance_micros_monthly, free_allowance_plans, free_allowance_reset) and the holding benefit (holding_threshold_mto, holding_multiplier, holding_verification_cadence). holding_benefit_open is always there, false until the holding benefit opens.
Read from hips.com on 2026-11-13 and abridged, the model id left as a placeholder and prompt_cache as the catalog answers it now.
Abridged answer
{ "data": { "revision": 131, "published_at": "2026-11-13T15:09:26.683968Z", "prices_updated_at": "2026-11-11T00:00:00Z", "models": [ { "id": "<model id>", "type": "language", "status": "live", "billing": { "unit": "tokens", "kinds": ["language_input", "language_input_cached", "language_output"] }, "pools": [ { "pool": "certified", "lane": "shared", "engine": "llamacpp", "precision": { "name": "Q4_K_M", "level": "int4" } }, { "pool": "community", "lane": "shared", "engine": "llamacpp", "precision": { "name": "Q4_K_M", "level": "int4" } } ], "batch": { "offered": false, "reason_code": null }, "prompt_cache": { "certified": false, "community": false }, "prices_published": true } ], "prices": [ { "kind": "language_input", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 137200, "batch": false, "pricing_mode": "market", "effective_from": "2026-11-10T02:41:42.059458Z", "scheduled": false }, { "kind": "language_input", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 68600, "batch": true, "pricing_mode": "market", "effective_from": "2026-11-10T02:41:42.059458Z", "scheduled": false }, { "kind": "language_input_cached", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 13720, "batch": false, "pricing_mode": "market", "effective_from": "2026-11-10T02:41:42.059458Z", "scheduled": false }, { "kind": "language_output", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 392000, "batch": false, "pricing_mode": "market", "effective_from": "2026-11-10T02:41:42.059458Z", "scheduled": false }, { "kind": "plan_fee", "model": null, "pool": null, "setting": null, "plan": "pro", "unit": "per_month", "amount_micros": 20000000, "batch": false, "pricing_mode": "market", "effective_from": "2026-10-26T00:00:00Z", "scheduled": false }, { "kind": "storage", "model": null, "pool": null, "setting": null, "plan": null, "unit": "per_gigabyte_month", "amount_micros": 100000, "batch": false, "pricing_mode": "market", "effective_from": "2026-10-27T00:00:00Z", "scheduled": false } ], "plans": [ { "plan": "pro", "terms": { "models": null, "features": ["router", "guardrails", "knowledge"], "limits": {} }, "effective_from": "2026-11-11T00:00:00Z", "scheduled": [] } ], "stable_schedules": [], "settings": { "holding_benefit_open": false } }}https://hips.com/api/models/{id}servedOne published model and its own price rows: the catalog's entry for that model, without the plans, the rows that are not a model's and the settings.
cURL
curl -s "https://hips.com/api/models/$HIPS_MODEL"Parts list: The answer's data
modelobjectThe model as the catalog lists it (The model fields that decide a price).
pricesarrayThe rows whose
modelis this one, in effect and announced, on demand and batch, each as in the catalog.revisionintegerThe catalog's revision.
published_atRFC 3339 timeWhen the catalog was last published.
{id} is a model's id as the catalog lists it, never a pinned id or an alias. An id the catalog does not list, a draft and a retired model answer 404 not_found.
Read from hips.com on 2026-11-13 and abridged, as the catalog's answer is.
Abridged answer
{ "data": { "model": { "id": "<model id>", "status": "live", "billing": { "unit": "tokens", "kinds": ["language_input", "language_input_cached", "language_output"] }, "batch": { "offered": true, "reason_code": null }, "prompt_cache": { "certified": false, "community": false }, "prices_published": true }, "prices": [ { "kind": "language_input", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 98000, "batch": false, "pricing_mode": "market", "effective_from": "2026-10-28T01:19:34.315366Z", "scheduled": false }, { "kind": "language_input", "model": "<model id>", "pool": "certified", "setting": null, "plan": null, "unit": "per_million_tokens", "amount_micros": 49000, "batch": true, "pricing_mode": "market", "effective_from": "2026-10-28T01:19:34.315366Z", "scheduled": false } ], "revision": 131, "published_at": "2026-11-13T15:12:52.663259Z" }}https://hips.com/api/prices/historyservedHow each published on-demand rate has moved since its model was listed: one series per model, pool, kind and setting, each rate with the moment it took effect.
curl -s https://hips.com/api/prices/historyParts list: The query
modelOptionalA model's
id: that model's series alone, with every point.sinceOptional, with modelThe start of a UTC day, written
YYYY-MM-DDT00:00:00Zand in no other form: each series then begins at the rate in effect at that moment.
Without model, each series carries its last 30 points; with it, every point, up to the newest 200 of a series. Any other parameter, a parameter given twice, a since in another form or without model, and a model outside the form of a model id (lowercase letters, digits, dots and hyphens) answer 400 invalid_request with the parameter in param; a model the catalog does not list answers 404 not_found.
Parts list: The answer's data
revisionintegerThe catalog's revision.
prices_updated_atRFC 3339 time or nullAs in the catalog.
seriesarrayOne series per model, pool, kind and setting.
Parts list: Each series
model, pool, kind, setting, unitAs on a price row;
poolnull on a series that applies in every pool the model is offered in.currentobjectThe rate in effect now:
effective_fromandamount_micros.previousobject or nullThe rate it replaced, or null for a first rate.
change_bpsinteger or nullThe change from
previoustocurrentin basis points ofprevious, rounded toward zero: 40000 is a rise of 400%, -2500 a fall of 25%. Null without a previous rate, or from a rate of zero.pointsarrayEach rate and the moment it took effect, oldest first. A rate that repeats the one before it is left out.
- The series cover the on-demand rates of the kinds each model is billed in (
language_input,language_input_cached,language_output,classification, andmedia_unitper setting) that took effect and were never canceled. A series begins when its model was listed: the rate in effect then is shown from that moment. current,previousandchange_bpsare the series' own, whateversinceor the cut to 30 points leaves inpoints.- Batch rows, announced rows, plans, storage and discounts have no history here; an announced change is in the catalog's
priceswithscheduledtrue.
Read from hips.com on 2026-11-13 and abridged, the model id left as a placeholder.
Abridged answer
{ "data": { "revision": 131, "prices_updated_at": "2026-11-11T00:00:00Z", "series": [ { "model": "<model id>", "pool": "certified", "kind": "language_input", "setting": null, "unit": "per_million_tokens", "current": { "effective_from": "2026-11-10T00:34:59.694295Z", "amount_micros": 196000 }, "previous": { "effective_from": "2026-11-08T21:42:17.004751Z", "amount_micros": 39200 }, "change_bps": 40000, "points": [ { "effective_from": "2026-11-08T21:42:17.004751Z", "amount_micros": 39200 }, { "effective_from": "2026-11-10T00:34:59.694295Z", "amount_micros": 196000 } ] } ] }}HIPS's machine-readable price feed: one document per model and pool, each with the model id that runs it in that pool, the pool's rates in force and what the pool serves for the model. Three feeds on https://ai.hips.com, the API's host, with no key:
/v1/feeds/certified/modelsservedThe Certified pool's offers, each under <model id>:certified, on https://ai.hips.com, with no key.
/v1/feeds/community/modelsservedThe Community pool's offers, each under <model id>:community, on https://ai.hips.com, with no key.
/v1/feeds/modelsservedBoth pools in one feed: a model's Certified offer at the default tier and its Community offer as service_tier: "flex" under the same id, on https://ai.hips.com, with no key.
cURL
curl -s --compressed https://ai.hips.com/v1/feeds/community/models- Each answers
{"data": [...]}, one document per offer, read from the catalog the API prices and bills every request by: a feed never lists a rate a request is not charged at that moment. Cache-Control: public, max-age=60and anETag; a matchingIf-None-Matchanswers304with no body. Sent withAccept-Encoding: gzip, the answer is gzipped, with its ownETagandVary: accept-encoding. A feed is made again when the catalog changes or an announced price takes effect.
- A language model that is published and offered with its rates (
statuslive), in each pool that serves it and has its own input and output rates in force there. - Never a model in preview, without a price in the pool, deprecated or retired, and never an announced price or a batch rate: the rates in force when the feed is made.
| Field | What HIPS writes |
|---|---|
schema_version | 2.4, the document format's version |
id | The model id that runs the model in the feed's pool, <model id>:certified or <model id>:community (A pool in the model id) |
name | The model's display name and its pool, as (Certified pool) or (Community pool) |
service_tier | flex on the combined feed's Community offers; left out everywhere else (The combined feed) |
hugging_face_id | The Hugging Face repository the model was added from, where the catalog shows it publicly |
quantization | The precision the pool serves the current version at, only where every kind of supply the pool serves it on states the same one: a GGUF's Q4_K_M is int4, Q8_0 int8, MXFP4 mxfp4, F16 fp16, BF16 bf16; left out wherever it is not recorded or differs |
input_modalities | Text, with max_context_length (the model's context in tokens) and its price; and images: inline (base64) images of image/png, image/jpeg and image/gif, with image/webp where the Certified pool's own servers take them, where the pool takes images, else an empty sources |
output_modalities | Text, with max_length (the model's output limit), streaming: true, supported_parameters and its price |
- Field
schema_version- What HIPS writes
2.4, the document format's version
- Field
id- What HIPS writes
- The model id that runs the model in the feed's pool,
<model id>:certifiedor<model id>:community(A pool in the model id)
- Field
name- What HIPS writes
- The model's display name and its pool, as
(Certified pool)or(Community pool)
- Field
service_tier- What HIPS writes
flexon the combined feed's Community offers; left out everywhere else (The combined feed)
- Field
hugging_face_id- What HIPS writes
- The Hugging Face repository the model was added from, where the catalog shows it publicly
- Field
quantization- What HIPS writes
- The precision the pool serves the current version at, only where every kind of supply the pool serves it on states the same one: a GGUF's
Q4_K_Misint4,Q8_0int8, MXFP4mxfp4,F16fp16,BF16bf16; left out wherever it is not recorded or differs
- Field
input_modalities- What HIPS writes
- Text, with
max_context_length(the model's context in tokens) and its price; and images: inline (base64) images ofimage/png,image/jpegandimage/gif, withimage/webpwhere the Certified pool's own servers take them, where the pool takes images, else an emptysources
- Field
output_modalities- What HIPS writes
- Text, with
max_length(the model's output limit),streaming: true,supported_parametersand its price
prompt and completion are the pool's input and output rates in force. cached_prompt is listed only where the pool bills cached input, which no pool does yet (Prompt caching). Each cost_usd is the rate in US dollars per token as an exact decimal string: the catalog's amount_micros per million tokens divided by 1,000,000,000,000, nothing rounded (A worked conversion). A price HIPS does not bill is left out, never written as zero.
supported_parameters names the sampling parameters of chat completions, always, each with its usual range: max_tokens (up to the model's output limit), temperature, top_p, stop (16 at most), seed, presence_penalty and frequency_penalty. A feature's parameters are listed only where the pool serves it for the model: tools and tool_choice where its tool calls are validated or its template takes them, parallel_tool_calls, response_format for json_object, structured_outputs for json_schema, logprobs and top_logprobs, logit_bias, and reasoning.
Note
The format is strict: a document holds only the fields its schema names, and HIPS's tests hold every document to them, so a reader that refuses an unknown field never refuses one of these.
For a reader that takes a single feed, GET /v1/feeds/models holds both pools: each model's Certified offer at the default tier under <model id>:certified, and its Community offer under that same id with service_tier: "flex". A model offered in the Community pool alone is listed at the default tier under <model id>:community.
- A chat completion that names a
:certifiedid withservice_tier: "flex"runs in the Community pool, exactly as the:communityid does, at the Community pool's rates and under its terms: any caller that sends it gets the Community pool.service_tieris not passed to the model. - The answer says the tier it was served at:
"service_tier": "flex"at the top level of the whole answer and of every streamed chunk. - Any other
service_tier, orflexon any other model id, a batch file's lines included, is a 400unsupported_parameter, as before. - A reader that shows a
flexoffer only while the default offer of its id is shown shows the combined feed's Community offer of a model only while its Certified offer is.
Abridged, its values left as placeholders:
One document
{ "data": [ { "schema_version": "2.4", "id": "<model id>:community", "name": "<display name> (Community pool)", "hugging_face_id": "<owner>/<name>", "quantization": "int4", "input_modalities": [ { "type": "text", "supported_inputs": { "max_context_length": { "value": "<the model's context>", "unit": "token" } }, "pricing": [{ "type": "prompt", "unit": "token", "cost_usd": "<rate in force, per token>" }] }, { "type": "image", "supported_inputs": { "sources": { "type": "enum", "values": [] } } } ], "output_modalities": [ { "type": "text", "max_length": { "value": "<the model's output limit>", "unit": "token" }, "streaming": true, "supported_parameters": { "max_tokens": { "type": "integer", "min": 1, "max": "<the model's output limit>", "unit": "token" }, "temperature": { "type": "range", "min": 0, "max": 2 }, "top_p": { "type": "range", "min": 0, "max": 1 }, "stop": { "type": "array", "max_items": 16 }, "seed": { "type": "integer" }, "presence_penalty": { "type": "range", "min": -2, "max": 2 }, "frequency_penalty": { "type": "range", "min": -2, "max": 2 }, "tools": { "type": "boolean" }, "tool_choice": { "type": "boolean" } }, "pricing": [{ "type": "completion", "unit": "token", "cost_usd": "<rate in force, per token>" }] } ] } ]}An error answers {"error": {"code", "message", "param", "request_id"}} with Cache-Control: no-store, request_id present where the platform behind the site answered it.
| Status | Code | When | Retry |
|---|---|---|---|
| 400 | invalid_request | The price history's query is out of shape; param names the parameter | No: correct the query |
| 404 | not_found | No published model has this id: one model's route, or the history's model | No |
| 502 | unreadable | The catalog's answer could not be read | Yes, later |
| 503 | not_connected | The site is not connected to the catalog | Yes, later |
| Another | The platform's own code | A failure behind the site, passed through with its status | Yes, later, for a 5xx |
- Status
- 400
- Code
invalid_request- When
- The price history's query is out of shape;
paramnames the parameter - Retry
- No: correct the query
- Status
- 404
- Code
not_found- When
- No published model has this id: one model's route, or the history's
model - Retry
- No
- Status
- 502
- Code
unreadable- When
- The catalog's answer could not be read
- Retry
- Yes, later
- Status
- 503
- Code
not_connected- When
- The site is not connected to the catalog
- Retry
- Yes, later
- Status
- Another
- Code
- The platform's own code
- When
- A failure behind the site, passed through with its status
- Retry
- Yes, later, for a 5xx
Related sheets
- Sheet
- 18 of 20
- Detail
- G
- Updated
- 2026-11-15