​
Source of truth

Use these base settings for all examples unless a page says otherwise:
ItemValue
Raw API basehttps://llm.ai-nebula.com
OpenAI SDK basehttps://llm.ai-nebula.com/v1
Auth headerAuthorization: Bearer <NEBULA_API_KEY>
Current availabilityRun GET /v1/models with your API key
Model IDs can share a provider name while requiring different endpoints or request parameters. Treat the model catalog and this routing table as the first step before copying an example.

​
Endpoint routing matrix

CapabilityModel familyRecommended endpointPayload shapeNotes
Text chatOpenAI-compatible chat models, Qwen, GLM, DeepSeek, Kimi, Grok, many Gemini/Claude models verified for chatPOST /v1/chat/completionsOpenAI ChatUse a minimal payload first. Optional params such as temperature and max_tokens are model-specific.
Text reasoningGPT-5, GPT-5 Pro, o-seriesPOST /v1/responsesOpenAI ResponsesRequired/preferred for GPT-5 pro family and reasoning controls. Use max_output_tokens.
Claude nativeClaude modelsPOST /v1/messagesAnthropic MessagesPreferred for Claude-native clients and full Anthropic features. Some Claude models reject chat temperature.
Claude token countClaude modelsPOST /v1/messages/count_tokensAnthropic token countEstimates input tokens before a Messages request.
Gemini native textGemini text modelsPOST /v1beta/models/{model}:generateContentGemini contents[].parts[]Use for Gemini-specific fields, native streaming, and provider-native compatibility.
OpenAI embeddingstext-embedding-3-small, text-embedding-3-large, text-embedding-ada-002POST /v1/embeddingsOpenAI EmbeddingsVerified working embedding route.
Gemini embeddingsgemini-embedding-001Not recommended until verifiedGemini native embeddingLatest audit returned HTTP 400 for the documented native shape. Use /v1/embeddings models for now.
RerankRerank models available to your keyPOST /v1/rerankNebula rerankConfirm model availability with GET /v1/models.
Image generationGPT Image, Gemini image / Nano Banana, Doubao Seedream/Seededit, Qwen Image, Imagen, Minimax imagePOST /v1/images/generationsProvider-specific image bodySame endpoint, different body per provider. Do not reuse one model family’s payload for another.
Video generationSora, Veo, Wanxiang, Seedance, HappyHorse, Minimax video where availablePOST /v1/video/generationsProvider-specific async bodySubmit task, poll GET /v1/video/generations/{task_id}. i2v/r2v modes require media inputs.
GPT Realtime v1gpt-realtime, gpt-realtime-miniwss://llm.ai-nebula.com/v1/realtime?model={model}OpenAI Realtime v1 eventsSession update uses session.modalities.
GPT Realtime v2gpt-realtime-2wss://llm.ai-nebula.com/v1/realtime?model=gpt-realtime-2OpenAI Realtime v2 eventsSession update uses session.type: "realtime"; do not send v1 modalities.
Gemini Livegemini-live-2.5-flash-native-audiowss://llm.ai-nebula.com/ws/v1beta/models/{model}/liveStreamGemini Live WebSocketSend setup first. Verify live behavior with the tested scripts before production use.
Audio speechgpt-4o-mini-ttsPOST /v1/audio/speechOpenAI audio speechReturns binary audio.

​
Safe starting payloads

​
Chat Completions

{
  "model": "qwen-plus",
  "messages": [
    { "role": "user", "content": "Reply with exactly: ok" }
  ]
}

​
Responses API

{
  "model": "gpt-5.2",
  "max_output_tokens": 64,
  "input": [
    { "role": "user", "content": "Reply with exactly: ok" }
  ]
}

​
Claude Messages

{
  "model": "claude-sonnet-4-5-20250929",
  "max_tokens": 64,
  "messages": [
    { "role": "user", "content": "Reply with exactly: ok" }
  ]
}

​
Gemini Native

{
  "contents": [
    {
      "role": "user",
      "parts": [{ "text": "Reply with exactly: ok" }]
    }
  ],
  "generationConfig": { "maxOutputTokens": 64 }
}

​
Rules of thumb

  • Check GET /v1/models before using a model ID.
  • Start with the minimal payload, then add optional parameters one by one.
  • Do not assume temperature, max_tokens, media fields, or search tools work across all providers.
  • For generated media, follow the provider-specific request body on the image or video page.
  • For WebSocket APIs, use a WebSocket client/Postman WebSocket mode, not normal HTTP send.