Skip to main content

Available Models

Every model below uses the same prefixed ID, such as openai/gpt-6-luna or perplexity/glm-5.3, in Chat Completions, Responses, and Messages. Context window and max output are in tokens; the context window covers input and output together. Rates are USD per 1 million tokens. A rate with a context-length range, such as (≤272k), changes when the input is above that range. The Usage tier column is the minimum usage tier that can call the model.

Anthropic

Claude Fable, Opus, Sonnet, and Haiku. Fable models are not available to organizations with zero data retention enabled.

Access by Usage Tier

A request for a model whose minimum usage tier is above your API organization’s tier returns HTTP 402. Organizations with zero data retention (ZDR) enabled cannot call the Claude Fable models; those requests return HTTP 403. The FAQ lists both messages. See Privacy & Security for how ZDR works.

Cached and Reasoning Tokens

Cache reads are billed at the cache-read rate. In Chat Completions and Responses, usage counts them inside the total input tokens and also reports them in cached_tokens; in Messages, input_tokens excludes them and cache_read_input_tokens reports them. Anthropic models and the GPT-6.1, GPT-6, and GPT-5.6 models bill tokens written to the cache once, at 1.25 times the model’s input rate. Chat Completions and Responses report them in cache_write_tokens; Messages reports them in cache_creation_input_tokens. No other model charges a cache-write premium. Anthropic models also accept a one-hour cache: set prompt_cache_options.ttl to "1h" on /chat/completions and /responses, or cache_control.ttl to "1h" on /messages. One-hour cache writes are billed at 2 times the model’s input rate. /messages reports them in cache_creation.ephemeral_1h_input_tokens; the other formats include them in cache_write_tokens. OpenAI models cache a repeated prefix on their own; Anthropic models cache only up to a cache breakpoint you mark. Reasoning tokens are billed at the output rate. The FAQ shows how to set caching and reasoning for each model family and where each format reports the tokens.

Estimate Your Cost

Model Routing

perplexity/auto is not a model in the tables. It picks a model for each request from the default policy or from routes you define, and you are billed at the rate of the model that answered. See Routing & Reliability.

List Models with the API

The response lists the models your usage tier can call, sorted by ID, with base token rates. It omits context windows, max output, and perplexity/auto. It does not apply zero data retention or organization model settings, so a ZDR organization still sees the Claude Fable models. Requesting a model ID that is not in the catalog returns HTTP 404. /responses sets error.code to model_not_found and param to model. /messages returns its native envelope with error.type set to not_found_error. On /responses:

Next Steps

Quickstart

Make your first Orchestrator API call.

Routing & Reliability

Model routing with perplexity/auto, deployment selection, and failover.

FAQ

Access, base URLs, billing, caching, requesting a model, and errors.