Developers

API

Models

GET /v1/models

Returns the list of models available through the Nebo API. The response follows the OpenAI models list format.

Request

curl https://janus.neboloop.com/v1/models \
  -H "Authorization: Bearer $NEBO_TOKEN"

Response

{
  "object": "list",
  "data": [
    {
      "id": "deepseek.v3.2",
      "object": "model",
      "created": 1712100000,
      "owned_by": "bedrock"
    },
    {
      "id": "gpt-5.2",
      "object": "model",
      "created": 1712100000,
      "owned_by": "openai"
    },
    {
      "id": "gpt-5-nano",
      "object": "model",
      "created": 1712100000,
      "owned_by": "openai"
    },
    {
      "id": "gpt-5-mini",
      "object": "model",
      "created": 1712100000,
      "owned_by": "openai"
    },
    {
      "id": "qwen.qwen3-32b",
      "object": "model",
      "created": 1712100000,
      "owned_by": "bedrock"
    }
  ]
}

The owned_by field indicates which provider serves the model.

Smart Routing vs Direct Selection

When you send a chat completion request, you choose how the model is selected:

Smart Routing

Set model to "auto" or leave it empty. Nebo's routing engine picks the best model for your message based on:

  • Content analysis — the complexity of your prompt (ML-based scoring)
  • Keyword rules — specific patterns that route to specialized models
  • Plan tier — your subscription determines which models are available
  • Cost optimization — cheaper models for simple tasks, premium models for complex reasoning
{"model": "auto", "messages": [...]}

This is the recommended approach for most use cases. You get optimal cost-performance without managing model selection yourself.

Direct Selection

Set model to a specific model ID from the list above. Your request routes directly to that model with no routing logic applied.

{"model": "deepseek.v3.2", "messages": [...]}

Use direct selection when you need deterministic model behavior — for example, testing, benchmarking, or workflows that depend on a specific model's capabilities.

Note: Direct model selection bypasses tier gating. Your request will be processed regardless of your plan tier, but it still consumes your usage budget. Expensive models consume budget faster.

Model Details

Model Provider Strengths Cost Tier
deepseek.v3.2 Bedrock Fast coding and reasoning, large context (164K) Low
qwen.qwen3-32b Bedrock General purpose, large context (131K) Lowest
gpt-5.2 OpenAI Complex reasoning, 400K context High
gpt-5-nano OpenAI Quick answers, trivial queries, 400K context Lowest
gpt-5-mini OpenAI Balanced performance, 400K context Low

Model availability may change. Always call GET /v1/models for the current list.

Embeddings Models

Embedding models are separate from chat models. See Embeddings for available embedding models.

Model Maps To Dimensions
neboloop/nebo-embed-small text-embedding-3-small 1536
neboloop/nebo-embed-large text-embedding-3-large 3072

How Budget Is Affected

Different models consume budget at different rates. When you use smart routing, Nebo automatically selects the most efficient model for each request. Advanced reasoning models consume more budget per request than lightweight models.

See Usage & Limits for budget details.

Next Steps

Chat Completions Embeddings