> ## Documentation Index
> Fetch the complete documentation index at: https://openrouter.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# PublicZdrEndpoint

An endpoint whose data policy does not retain prompts by default

## Fields

| Field | Type | Required | Description | Example |
| - | - | - | - | - |
| `context_length` | *int* | ✔️ | N/A | |
| `latency_last_30m` | [Nullable\[components.PercentileStats\]](../components/percentilestats.mdx) | ✔️ | Latency percentiles in milliseconds over the last 30 minutes. Latency measures time to first token. Only visible when authenticated with an API key or cookie; returns null for unauthenticated requests. | \{ "p50": 25.5, "p75": 35.2, "p90": 48.7, "p99": 85.3 } |
| `max_completion_tokens` | *Nullable\[int]* | ✔️ | Maximum completion tokens for this endpoint. Input and output tokens share the context window, so the effective maximum output for a request is further limited by the context remaining after input tokens. | |
| `max_prompt_tokens` | *Nullable\[int]* | ✔️ | N/A | |
| `model_id` | *str* | ✔️ | The unique identifier for the model (permaslug) | openai/gpt-4 |
| `model_name` | *str* | ✔️ | N/A | |
| `name` | *str* | ✔️ | N/A | |
| `native_tools` | Dict\[str, [components.NativeTools](../components/nativetools.mdx)] | ✔️ | The server tools this endpoint accepts as the provider's own built-in tool (`engine: "native"`) instead of an OpenRouter engine, keyed by canonical `openrouter:*` name. Each value names the provider tool type the request is translated to. Where that tool runs (provider-side, or returned to the client as with Anthropic bash) is documented per tool. Empty when the provider has none. | \{ "openrouter:web\_search": \{ "type": "web\_search\_20260209" } } |
| `perf_last_30m_by_workload` | [Optional\[components.PerfLast30mByWorkload\]](../components/perflast30mbyworkload.mdx) | ➖ | Endpoint performance over the last 30 minutes, keyed by the kind of request served (e.g. `text_generation`, `image_generation`). Additive to the legacy singular latency and throughput fields; image and video generation report end-to-end latency. Only visible when authenticated with an API key or cookie. | |
| `pricing` | [components.Pricing](../components/pricing.mdx) | ✔️ | N/A | |
| `provider_name` | [components.ProviderName](../components/providername.mdx) | ✔️ | N/A | OpenAI |
| `quantization` | [Nullable\[components.Quantization\]](../components/quantization.mdx) | ✔️ | N/A | fp16 |
| `service_tier` | [Nullable\[components.PublicServiceTier\]](../components/publicservicetier.mdx) | ✔️ | The service tier this endpoint sells, or null for a standard endpoint. The same value whether the provider selects the tier with a request parameter or with a dedicated model id. Always the canonical tier: priority covers the fast request alias and the turbo, highspeed, and prime provider model-id spellings, none of which this field reports. | priority |
| `status` | [Optional\[components.EndpointStatus\]](../components/endpointstatus.mdx) | ➖ | N/A | 0 |
| `supported_parameters` | List\[[components.Parameter](../components/parameter.mdx)] | ✔️ | N/A | |
| `supports_diarization` | *bool* | ✔️ | Whether this STT endpoint can label each transcript segment with a speaker (speaker diarization). False when the endpoint has not declared support. | |
| `supports_image_reference` | *Optional\[bool]* | ➖ | Whether this TTS endpoint accepts an `image_url` reference describing the desired voice. Requests carrying an image reference are only routed to endpoints where this is true. | |
| `supports_implicit_caching` | *bool* | ✔️ | N/A | |
| `supports_multiple_audio_references` | *Optional\[bool]* | ➖ | Whether this TTS endpoint accepts more than one `input_audio` reference clip per request. Requests carrying several clips are only routed to endpoints where this is true. | |
| `supports_tool_choice` | [components.ToolChoiceSupport](../components/toolchoicesupport.mdx) | ✔️ | Per-variant `tool_choice` support. `tool_choice` in `supported_parameters` only says the parameter is accepted; these flags say which of its values passed testing. | \{ "auto": true, "function": true, "none": true, "required": true } |
| `supports_voice_cloning` | *Optional\[bool]* | ➖ | Whether this TTS endpoint accepts inline reference audio (`input_references`) for stateless voice cloning. Requests carrying reference audio are only routed to endpoints where this is true. | |
| `tag` | *str* | ✔️ | N/A | |
| `throughput_last_30m` | [Nullable\[components.PercentileStats\]](../components/percentilestats.mdx) | ✔️ | N/A | \{ "p50": 25.5, "p75": 35.2, "p90": 48.7, "p99": 85.3 } |
| `uptime_last_1d` | *Nullable\[float]* | ✔️ | Uptime percentage over the last day: the share of minutes in which at least 80% of provider attempts succeeded, counting only minutes with 10 or more attempts. Rate-limited and caller-caused failures are excluded. Null when no minute had enough traffic. | |
| `uptime_last_30m` | *Nullable\[float]* | ✔️ | N/A | |
| `uptime_last_5m` | *Nullable\[float]* | ✔️ | Uptime percentage over the last 5 minutes: the share of minutes in which at least 80% of provider attempts succeeded, counting only minutes with 10 or more attempts. Rate-limited and caller-caused failures are excluded. Null when no minute had enough traffic. | |
| `safety_retention` | *bool* | ✔️ | Whether the endpoint uses Safety Retention: prompts are not retained by default, but the provider may retain prompts its safety classifiers flag. A ZDR-enforced request reaches these endpoints only when the account allows Safety Retention and no applied guardrail disallows it. | false |