Models

The model field takes the exact model name. The full list with prices and cache support is on the catalog page.

• IDs use vendor/model (OpenRouter-style), e.g. anthropic/claude-opus-5; claude-opus-5 is accepted as an input alias with identical billing.
• model:alias pins a channel, e.g. openai/gpt-5.6-sol:yellow. GET /v1/models lists one ID per model; ?variants=1 returns the full list.

Most channels are marketplace books: many sellers offer the same model at different prices. Your request fills at the cheapest offer that takes it, and you are billed what that fill cost plus our commission. The per-token rate is therefore a range, not a single number — it depends on which rung of the book filled.

GET /api/catalog and GET /v1/pricing publish that range per model (pricing) and per channel (channels[]), read from the same ladder GET /api/model-ladder?model=<model> serves with its offer counts.

"pricing": {
  "basis": "fill",
  "unit": "USD per 1M tokens",
  "priceFrom":    { "input": 0.465, "output": 2.325 },
  "priceTypical": { "input": 0.5,   "output": 2.5 },
  "priceTo":      { "input": 1.5,   "output": 7.5 },
  "minChargePerRequest": 0.0000125,
  "ladder": "/api/model-ladder?model=claude-opus-5.5",
  "ladderUpdatedAt": "2026-09-25T08:45:01.522Z",
  "ladderStale": false
},
"channels": [
  { "address": "claude-opus-5.5:claude-code", "isDefault": true,
    "inputUsd": 0.465, "outputUsd": 2.325,
    "basis": "fill", "pool": "cc/claude-opus-5-5",
    "priceFrom": { "input": 0.465, "output": 2.325 },
    "priceTypical": { "input": 0.5, "output": 2.5 },
    "priceTo": { "input": 1.5, "output": 7.5 },
    "maxPriceCeiling": { "input": 0.6, "output": 3 },
    "minChargePerRequest": 0.0000125 }
]

Most models have several public variants. The default variant is chosen automatically (priority follows your account or key settings). To pin a specific variant for a request, add its alias after a colon to the model name:

"model": "gpt-6.1-sol"           // default variant (auto)
"model": "gpt-6.1-sol:yellow"    // explicit public variant
"model": "claude-opus-5.5"     // use an id from the catalog

Example request with an explicit variant:

curl https://getvibeapi.com/v1/chat/completions \
  -H "Authorization: Bearer $GETVIBE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4.8:yellow",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'
# :yellow — pin a public variant for this request

Use only aliases from the catalog. Available suffixes:

You can fetch the full list of available model ids programmatically:

curl https://getvibeapi.com/v1/models \
  -H "Authorization: Bearer $GETVIBE_API_KEY"

Besides id/object/created/owned_by, every entry carries type (chat, embedding, rerank, image, video, audio, music, 3d), architecture.input_modalities / output_modalities and endpoints — where to send a request for that model.

• The default list holds chat, embedding and rerank models only, so an IDE model picker never offers a model that cannot chat.
• Media models (image, video, audio, music, 3D) are listed on request: ?type=image (or a list: ?type=image,video), ?output_modalities=image; ?modality= and ?category= are synonyms of type.
• Everything at once: ?all=true, ?type=all or GET /v1/models/all. Media is listed together with its variants (model:alias) — exactly the id to send.
• GET /v1/models/{id} returns one model; an unknown id is a 404 with code: "model_not_found".

curl "https://getvibeapi.com/v1/models?type=image,video" \
  -H "Authorization: Bearer $GETVIBE_API_KEY"

Every model added to the catalog and our what's-new posts are published in an open feed — no key needed. Read it from an RSS reader or from an agent, so it can change its approach as models change. The feed holds the whole history, page by page, and every event carries what an agent needs to call the model right away.

• /rss.xml (alias /feed.xml) is RSS 2.0; /feed.json is JSON Feed 1.1 with a feed-level _meta block and a _catalog block per event.
• Pages: 50 events each, ?limit= up to 100, newest first. The next (older) page is next_url in JSON Feed and <atom:link rel="next"> in RSS (RFC 5005, with rel="first" and rel="previous" too). The links carry a cursor, so a model published while you page through never shifts an event you already read. Use the links as returned.
• Filters: ?kind=text|image|video|audio|music|3d|news and ?since=YYYY-MM-DD — only events from that UTC day on, inclusive. Filters work together with paging and are kept in the links.
• A model_added event carries the title, the date the model appeared in the catalog, a link to the model page, a description and, in _catalog: the public ids to send as model (no :alias route spellings and no supplier names), the endpoints, input and output modalities, context window and max output, capabilities (tools, JSON / structured output, vision, reasoning and its levels, caching), the public starting price in USD with its unit, for media every request parameter with its allowed values, a ready request (curl and JSON) and a docs link. A news event is a what's-new post.
• An event's guid never changes. The feed has no recorded generation results, only an example request.
• Poll at most every few minutes: every page has its own ETag and Last-Modified; send If-None-Match / If-Modified-Since and an unchanged page answers 304 with no body. To fetch only what is new, remember the date of the last event you processed and request ?since= with it.

curl -sS "https://getvibeapi.com/feed.json?since=2026-10-01&limit=100"
curl -sS -i https://getvibeapi.com/feed.json -H 'If-None-Match: "<ETag from the previous response>"'   # -> 304