Models & pricing

This page is for developers choosing a callable model and estimating the likely cost before integration.

OpenAI-compatible endpoints accept either a complete author/model ID or a bare model name. The gateway resolves exact matches from its alias table. For example, gpt-5.4-nano resolves to openai/gpt-5.4-nano, deepseek-v4-flash to deepseek/deepseek-v4-flash, qwen3.5-flash to qwen/qwen3.5-flash-02-23, and claude-haiku-4.5 to anthropic/claude-haiku-4.5. A bare name absent from the alias table is treated as openai/<name>.

Prefer the complete ID in production and inspect the response model field to confirm the resolved ID. Variants are not bare aliases: always write the complete author/model:free or author/model:batch value.

Anthropic and Gemini native endpoints use different namespaces. Anthropic Messages uses native hyphenated names, such as the verified claude-haiku-4-5. The gateway maps a few dotted aliases (claude-opus-4.8 resolves to claude-opus-4-8), but coverage is partial: claude-haiku-4.5 returns 404. OpenAI-compatible IDs such as anthropic/claude-haiku-4.5 are never valid there. Use public GET /anthropic/v1/models for its current list. Gemini accepts Google-native names, such as the verified gemini-3.1-flash-lite; use public GET /v1beta/models for that protocol.

Base URLs and primary endpoints#

OpenAI-compatible calls use the base URL https://api.clssai.com/v1 and accept Bearer, x-api-key, or a ?key= query parameter. Prefer a header: a key in the URL leaks through referrers, proxy logs, and browser history. The Anthropic SDK base URL is https://api.clssai.com without /v1; calls accept x-api-key or Bearer and require anthropic-version: 2023-06-01. Gemini uses https://api.clssai.com/v1beta with x-goog-api-key or ?key= and does not accept Bearer authentication.

List callable model IDs#

GET /v1/models is public (no API key), supports ETag and If-None-Match, and returns the full reference record for every callable model. Catalog routes answer locally and are never forwarded upstream.

curl "https://api.clssai.com/v1/models?limit=1&q=gpt-5.4-nano"

{
  "object": "list",
  "data": [
    {
      "id": "openai/gpt-5.4-nano",
      "object": "model",
      "created": 1773748187,
      "owned_by": "openai",
      "canonical_slug": "openai/gpt-5.4-nano-20260317",
      "hugging_face_id": "",
      "name": "OpenAI: GPT-5.4 Nano",
      "description": "GPT-5.4 nano is the most lightweight and cost-efficient variant of the GPT-5.4 family, optimized for speed-critical and high-volume tasks. It supports text and image inputs and is designed for low-latency...",
      "context_length": 400000,
      "architecture": {
        "modality": "text+image+file->text",
        "input_modalities": ["file", "image", "text"],
        "output_modalities": ["text"],
        "tokenizer": "GPT",
        "instruct_type": null
      },
      "pricing": {
        "prompt": "0.0000002",
        "completion": "0.00000125",
        "web_search": "0.01",
        "input_cache_read": "0.00000002"
      },
      "top_provider": {
        "context_length": 400000,
        "max_completion_tokens": 128000,
        "is_moderated": true
      },
      "per_request_limits": null,
      "supported_parameters": [
        "include_reasoning",
        "max_completion_tokens",
        "max_tokens",
        "reasoning",
        "reasoning_effort",
        "response_format",
        "seed",
        "structured_outputs",
        "tool_choice",
        "tools"
      ],
      "default_parameters": {
        "temperature": null,
        "top_p": null,
        "top_k": null,
        "frequency_penalty": null,
        "presence_penalty": null,
        "repetition_penalty": null
      },
      "supported_voices": null,
      "knowledge_cutoff": "2025-08-31",
      "expiration_date": null,
      "links": {
        "details": "/v1/models/openai/gpt-5.4-nano"
      }
    }
  ],
  "total_count": 2,
  "links": {
    "next": "/v1/models?limit=1&q=gpt-5.4-nano&offset=1"
  }
}

The API directory is the authority for callable IDs. Reference context and token prices below are server-rendered from the same catalog snapshot used by the full model browser. Non-token-priced models show an em dash instead of a token price.

Response fields#

Each record in data carries the four OpenAI list fields plus the full reference metadata:

architecture contains modality (a input->output summary string), input_modalities, output_modalities, tokenizer, and instruct_type. Modalities are chosen from text, image, file, audio, video, embeddings, rerank, speech, transcription, and decisions.

Every pricing value is a decimal string in USD. prompt and completion are prices per token; keys such as web_search, input_cache_read, input_cache_write, and input_cache_write_1h are prices per action or per token of that class. A free model reports "0" for prompt and completion. Do not treat catalog prices as a final invoice; see Usage & billing.

top_provider contains context_length, max_completion_tokens, and is_moderated. Use top_provider.context_length and top_provider.max_completion_tokens when you need the serving route's limits rather than the catalog headline.

Query the directory#

All catalog list routes accept the same query parameters. Omitting every parameter returns the full list; pass limit to page.

Unknown names in output_modalities, input_modalities, or supported_parameters match nothing and return an empty data array with total_count: 0. Invalid limit, min_price, max_price, or sort values return 400 invalid_parameter and name the offending parameter.

curl "https://api.clssai.com/v1/models?limit=2&output_modalities=image&sort=name"

Top-level response fields are object (list), data, total_count, and links. total_count is the number of records matching the filter, not the page size. links.next is the path for the following page, or null when the page is the last one.

{
  "object": "list",
  "data": [],
  "total_count": 55,
  "links": {
    "next": "/v1/models?limit=2&output_modalities=image&sort=name&offset=2"
  }
}

Fetch one model#

GET /v1/models/{author}/{slug} returns one model object (the same fields as a data entry above) without a list wrapper. slug is the final ID segment and keeps any :free or :batch suffix. Unknown IDs return 404 not_found.

curl https://api.clssai.com/v1/models/openai/gpt-5.4-nano

{
  "id": "openai/gpt-5.4-nano",
  "object": "model",
  "created": 1773748187,
  "owned_by": "openai",
  "canonical_slug": "openai/gpt-5.4-nano-20260317",
  "name": "OpenAI: GPT-5.4 Nano",
  "context_length": 400000,
  "links": {
    "details": "/v1/models/openai/gpt-5.4-nano"
  }
}

Count models#

GET /v1/models/count returns the number of models matching no filter. It is the cheap way to detect a catalog change before downloading the list.

curl https://api.clssai.com/v1/models/count

{"data":{"count":622}}

Filter by output modality#

Three public shortcuts return the same rich records, pre-filtered by output modality. They accept every query parameter above.

curl "https://api.clssai.com/v1/embeddings/models?limit=1"

{
  "object": "list",
  "data": [
    {
      "id": "liquid/lfm-2.5-embedding-350m:free",
      "object": "model",
      "created": 1787077908,
      "owned_by": "liquid",
      "architecture": {
        "modality": "text->embeddings",
        "input_modalities": ["text"],
        "output_modalities": ["embeddings"]
      },
      "pricing": {
        "prompt": "0",
        "completion": "0"
      },
      "links": {
        "details": "/v1/models/liquid/lfm-2.5-embedding-350m:free"
      }
    }
  ],
  "total_count": 37,
  "links": {
    "next": "/v1/embeddings/models?limit=1&offset=1"
  }
}

Models by modality#

Common mistakes#

• Include /v1 in the OpenAI SDK base URL.
• Do not infer availability from a model detail URL; use GET /v1/models.
• Prefer a complete author/model ID in production, use the response model field to audit bare-name resolution, and keep :free or :batch in the complete ID.
• Do not send OpenAI-compatible IDs to native Anthropic or Gemini endpoints; use each protocol's public model list.
• Do not treat a displayed reference price as the final charge; see Usage & billing.
• Do not call inference endpoints from browser JavaScript.
• Use the Anthropic SDK base URL without /v1.
• Use cf-ray when tracing inference requests.
• Do not assume limit is optional for pagination; without it a request can return the entire directory.
• Treat pricing strings as decimal USD values and convert them explicitly; they are not floating-point JSON numbers.