Models

List and inspect every model your API key can call.

Last updated May 1, 2026

Cocobox exposes a curated set of models from Amazon Bedrock under friendly aliases like tambor, prisma, sonnet, and opus. The same alias works on both the OpenAI surface (/v1/chat/completions) and the Anthropic surface (/v1/messages).

List models

GET https://api.cocobox.io/v1/models

Returns the menu in the OpenAI shape:

{
  "object": "list",
  "data": [
    { "id": "tambor",  "object": "model", "owned_by": "cocobox", "created": 1735689600 },
    { "id": "haiku",   "object": "model", "owned_by": "cocobox", "created": 1735689600 },
    { "id": "sonnet",  "object": "model", "owned_by": "cocobox", "created": 1735689600 },
    { "id": "opus",    "object": "model", "owned_by": "cocobox", "created": 1735689600 }
  ]
}

To get the Anthropic shape with display names, send Accept: application/anthropic (or call from a client that already sets anthropic-version):

{
  "data": [
    {
      "id": "sonnet",
      "type": "model",
      "display_name": "Claude Sonnet 4.6",
      "created_at": "2025-01-01T00:00:00Z"
    }
  ],
  "has_more": false
}

The endpoint always returns only the models your key is allowed to call — allowedModels on the key filters this list.

Aliases vs full ids

Cocobox accepts both:

{ "model": "sonnet" }
{ "model": "global.anthropic.claude-sonnet-4-6" }

Aliases are stable over time. Full ids may change when Bedrock revs a model — when that happens we update the alias to point at the new full id and announce in the Changelog. Prefer aliases.

Catalog

For the complete table of aliases, full ids, credit costs, context windows, and recommended use-cases, see Reference → Supported models.

A short version:

Alias	Provider	Tier	Credits / req	Best for
`tambor`	Amazon Nova Lite	Cheap	0.33	Everyday SQL, autocomplete
`trueno`	Mistral 7B	Standard	1	Fast classifications, simple Q&A
`haiku`	Anthropic Haiku 4.5	Standard	1	Snappy answers with strong reasoning
`prisma`	Amazon Nova Pro	Standard	3	Long-context analysis
`sonnet`	Anthropic Sonnet 4.6	Standard	3	Complex SQL, multi-step reasoning
`sonnet-5`	Anthropic Sonnet 5	Standard	2	Newest Sonnet, faster and cheaper
`nova-premier`	Amazon Nova Premier	Pro	3	Hardest queries, deep reasoning
`deepseek-r1`	DeepSeek R1	Pro	2	Math-heavy / analytical
`opus`	Anthropic Opus 4.6	Premium	7	The most demanding work
`opus-4-7`	Anthropic Opus 4.7	Premium	7	Newest Opus, frontier reasoning
`opus-4-8`	Anthropic Opus 4.8	Premium	7	Latest Opus, frontier reasoning
`fable-5`	Anthropic Fable 5	Premium	10	Highest-tier model, hardest problems

Credits are deducted before the call goes to the provider. If your subscription pool is empty, the request returns 402 insufficient_credits immediately. See Credits & billing.

Choosing a model

For most SQL prompts start with tambor (0.33 credits — ~900 calls / month included) or haiku (1 credit, faster reasoning).
For complex joins, window functions, or multi-step refactors, sonnet or prisma (both 3 credits).
For whole-schema audits or long-running analytical tasks, use opus or nova-premier.

The Cocobox app picks intelligently for you by default; the API leaves the choice in your hands.