Inference

Models

List the models available to your project, their capabilities, limits and prices.

Models are addressed by a stable gateway ID such as claude-sonnet or gpt-5-mini. The gateway maps each ID to the provider's current upstream model, so your code does not change when the provider's model name does.

List models

GET/v1/models

Returns the models that can serve requests on this deployment right now. Any valid API key can call it; no scope is required. The endpoint is not rate limited.

Shell

curl https://www.hushcompute.xyz/v1/models \
  -H "Authorization: Bearer $HUSH_API_KEY"

200 OK

{
  "object": "list",
  "data": [
    {
      "id": "claude-sonnet",
      "object": "model",
      "created": 1781049600,
      "owned_by": "anthropic",
      "display_name": "Claude Sonnet 5",
      "status": "live",
      "privacy_tier": "standard",
      "capabilities": ["chat", "streaming", "tools", "reasoning"],
      "context_window": 1000000,
      "max_output_tokens": 32000,
      "pricing": {
        "unit": "micro_usd_per_million_tokens",
        "input": "2200000",
        "output": "11000000"
      }
    }
  ]
}

Model object

Field	Type	Description
id	string	The model ID to pass as model in requests.
object	string	Always model.
created	integer	Unix time the model was added to the catalog.
owned_by	string	Upstream provider: openai, anthropic, google, openrouter, or dev.
display_name	string	Human-readable name.
status	string	live, beta or preview. See Statuses.
privacy_tier	string	Routing privacy tier. Only standard is currently available.
capabilities	string[]	Features the model supports. See Capabilities.
context_window	integer	Maximum context length in tokens, as published by the provider.
max_output_tokens	integer	Gateway cap on output tokens per request. Larger max_completion_tokens values are reduced to this.
pricing	object	unit is micro_usd_per_million_tokens; input and output are integer µUSD prices as decimal strings.

Catalog

The platform catalog defines the models below. Whether each one is served depends on the providers configured on your deployment; GET /v1/models is the authoritative list.

Model ID	Provider	Status	Capabilities	Context	Max output	Sampling
claude-opus	Anthropic	Live	chatstreamingtoolsreasoning	1,000,000	32,000	Ignored
claude-sonnet	Anthropic	Live	chatstreamingtoolsreasoning	1,000,000	32,000	Ignored
claude-haiku	Anthropic	Live	chatstreamingtools	200,000	16,000	Yes
gpt-5	OpenAI	Live	chatstreamingtoolsreasoning	400,000	32,000	Ignored
gpt-5-mini	OpenAI	Live	chatstreamingtoolsreasoning	400,000	32,000	Ignored
gemini-2.5-pro	Google	Beta	chatstreamingreasoning	200,000	32,000	Yes
gemini-2.5-flash	Google	Beta	chatstreaming	1,000,000	32,000	Yes
llama-3.3-70b	OpenRouter	Preview	chatstreamingtools	128,000	16,000	Yes

Sampling indicates whether temperature and top_p are forwarded. For models marked Ignored the parameters are accepted and silently dropped. When max_completion_tokens is omitted, requests default to 4,096 output tokens or the model's cap, whichever is lower.

Capabilities

Capability	Meaning
chat	Accepts Chat Completions and Responses requests.
streaming	Supports stream: true.
tools	Accepts function tools. Sending tools to a model without this capability returns 400 invalid_request.
reasoning	The upstream model performs internal reasoning. Reasoning content is not returned.
json	The upstream model supports structured output. The gateway does not yet accept response_format; the field is ignored.
vision	Reserved. Image inputs are not yet supported by the gateway.

Statuses

Live Production model with list pricing verified by the operator.
Beta Available, but behaviour or pricing may change.
Preview Early access through a routing provider; expect changes.

Models whose provider is not configured are omitted from the list. dev-echo, a deterministic model for local testing, appears only on non-production deployments with the development provider enabled.

Pricing

Prices are integer µUSD per one million tokens and include the platform margin configured by the operator. Read them from GET /v1/models rather than hardcoding them. The cost of a request is computed separately for input and output and rounded up to the next µUSD:

Cost formula

cost = ceil(input_tokens  × input_price  / 1,000,000)
     + ceil(output_tokens × output_price / 1,000,000)

example: 480 input tokens at 1,100,000 µUSD/M  →    528 µUSD
         310 output tokens at 5,500,000 µUSD/M →  1,705 µUSD
         total                                  →  2,233 µUSD ($0.002233)

The example response above shows illustrative prices. Reservations, settlement and budgets are described in Usage.

Availability

An unknown model ID returns model_not_found (HTTP 404).
A catalog model that is unavailable, or whose provider is not configured on the deployment, returns service_unavailable (HTTP 503).
A model excluded by the project policy or the key's allowlist returns model_not_allowed (HTTP 403).