Models

GET /v1/models

List available models with their capabilities.

Response

Returns available models grouped by provider. Each entry includes a capabilities object describing what the model supports end-to-end through PickLayer’s request pipeline.


{
  "data": {
    "providers": {
      "openai": [{
        "provider": "openai",
        "model": "gpt-4o-2024-11-20",
        "modelString": "openai/gpt-4o-2024-11-20",
        "canonical_slug": "gpt-4o-2024-11-20",
        "model_kind": "chat",
        "capabilities": {
          "token_unit": "tokens",
          "supports_text": true,
          "supports_tools": true,
          "supports_vision": true,
          "supports_web_search": true,
          "supports_web_search_native": false,
          "web_search": { "mode": "platform_tool", "source": "derived", "supported": true },
          "supports_file_upload": true,
          "supports_json_mode": true,
          "supports_reasoning": false,
          "supports_system_prompt": true,
          "max_output_tokens": 16384,
          "context_window_tokens": 128000,
          "tokenization_hint": "openrouter:openai/gpt-4o-2024-11-20",
          "accepted_extensions": ["txt","md","csv","tsv","json","yaml","yml","xml","html","pdf","docx","xlsx","jpg","jpeg","png","gif","webp"],
          "accepted_mime_prefixes": ["text/","application/pdf","application/json","application/vnd.openxmlformats-officedocument.","image/"]
        }
      }]
    }
  }
}

Capabilities

Field	Type	Description
`supports_text`	boolean	Model accepts and returns text content.
`supports_vision`	boolean	Model accepts image inputs (`image_url`/`input_image`).
`supports_tools`	boolean	Model supports function/tool calling.
`supports_json_mode`	boolean	Model honors `response_format: { type: "json" }`.
`supports_reasoning`	boolean	Model produces reasoning traces (e.g., o1/o3, Claude extended thinking).
`supports_system_prompt`	boolean	Model accepts a system role / system instruction.
`supports_web_search`	boolean	Web search is available — either via PickLayer’s platform-tool implementation (any tool-capable model) or a provider-native search backend. Frontends should read this field directly to decide whether to surface a “Web Search” toggle.
`supports_web_search_native`	boolean	Provider exposes a native web-search tool (Anthropic Brave, Gemini Google Search, xAI Live Search, OpenAI native).
`web_search`	object	Detailed web-search metadata: `{ "mode": "native" \| "platform_tool", "source": "openrouter_api" \| "provider_table" \| "derived", "supported": true }`.
`supports_file_upload`	boolean	Model accepts file attachments (`input_file`, `image_url`, `input_audio`).
`accepted_extensions`	string[]	File extensions (no leading dot) the request pipeline can route end-to-end. Derived from runtime capability — not aspirational.
`accepted_mime_prefixes`	string[]	MIME prefixes the request pipeline can route end-to-end.
`context_window_tokens`	number	Maximum input context size.
`max_output_tokens`	number	Maximum output length the gateway will accept (capped against provider limits).
`tokenization_hint`	string	Hint for token counting (e.g., `"openrouter:openai/gpt-4o-2024-11-20"`); used by `/v1/messages/count_tokens` and the embedded tokenizer.
`token_unit`	string	`"tokens"` for chat models; `"seconds"` for STT/TTS.

`supports_web_search` derivation

Canonical-additive change. supports_web_search and the 11 sibling capability fields described here live on PickLayer’s extended capabilities object. The OpenAI-compatible flat-shape OpenAIModelObjectSchema exposed at the canonical OpenAI catalog surface is untouched — clients reading the canonical OpenAI shape see no change.

PickLayer’s platform web search works via tool-calling on any tool-capable model. The synthesis rule is:


// Explicit value wins; only derive when the field is undefined.
if (capabilities.supports_web_search === undefined) {
  capabilities.supports_web_search = canUsePlatformWebSearch(capabilities);
  // canUsePlatformWebSearch returns supports_tools === true
}

If a per-provider capability table sets supports_web_search: false explicitly for cost or policy reasons, that value is preserved. If a provider’s enrichment (e.g., OpenRouter’s API) reports supports_web_search: true directly, that wins. Only when the field is otherwise undefined does the gateway fall back to deriving from supports_tools.

This means native-vendor entries that don’t pass through OpenRouter enrichment (e.g., openai/gpt-4o-2024-11-20) now correctly advertise supports_web_search: true instead of leaving the field undefined.

Truthfulness invariant

accepted_extensions and accepted_mime_prefixes are derived from the runtime capability — i.e., the advertised list reflects what the request pipeline can actually process end-to-end. A request with an input_file whose MIME is not in the model’s advertised list returns 400 with a precise param path; no aspirational entries appear in the public list.

Backwards-compatibility note

Until clients ship a supports_web_search read path, they may infer it from supports_tools. Once they pick up the new field, prefer reading it directly: it preserves explicit per-model overrides that a supports_tools-only inference would miss.

Query parameters

Parameter	Description
`provider`	Comma-separated filter (e.g., `provider=openai,anthropic`).

/v1/models/available is an alias for the same response.