Narev lookup llm pricing

Look up LLM pricing

Reference for the only Narev pricing HTTP endpoints agents should use. Base URL: https://api.narev.ai. All are public (no Authorization: Bearer or NAREV_API_KEY).

Rates in price responses are USD per token (not per 1K or 1M). For committed snapshots, use update-llm-pricing.

Endpoints (exhaustive list)

Do not call GET /v1/models/pricing, GET /models/pricing, POST /models/pricing/calculate, POST /v1/models/pricing/calculate, or any other path not listed here.

Discovery

• GET https://api.narev.ai/v1/reference/providers — all providers (provider_id, name).
• GET https://api.narev.ai/v1/reference/providers/{provider_id} — one provider (base_url, pricing_url, logo_url, …). 404 if unknown.
• GET https://api.narev.ai/v1/reference/models — model IDs per provider. Query: provider_ids (comma-separated), page, page_size (max 1000).

Rates

• GET https://api.narev.ai/v1/price/{provider_id} — pricing for models on that provider. Query: model_id, page, page_size.
• GET https://api.narev.ai/v1/price/search — search pricing by model ID. Query: q, page, page_size.
• GET https://api.narev.ai/v1/find/cheapest/{model_id} — every provider that serves this model, sorted cheapest pricing.prompt first. Query: page, page_size.

Calculate

• POST https://api.narev.ai/v1/calculate — USD total for one call (modelId, provider, usage; optional subprovider, isByok, webSearchCount).

Typical flow

1. Which providers exist? → GET /v1/reference/providers
2. Which models exist? → GET /v1/reference/models (optionally provider_ids=openai,anthropic)
3. What are the token rates?
   • One provider: GET /v1/price/openai?model_id=gpt-4o
   • Search by name: GET /v1/price/search?q=gpt-4
   • Compare hosts for one model: GET /v1/find/cheapest/claude-sonnet-4-20250514
4. What did this call cost in USD? → POST /v1/calculate with token counts

Paginate price and reference list endpoints while page <= meta.total_pages.

Price response shape

GET /v1/price/* and GET /v1/find/cheapest/* return:

{
  "data": [
    {
      "model_id": "gpt-4o",
      "provider_id": "openai",
      "pricing": {
        "prompt": 0.0000025,
        "completion": 0.00001,
        "discount": 0,
        "request": 0,
        "web_search": 0,
        "input_cache_read": 0.00000125,
        "input_cache_write": 0,
        "internal_reasoning": 0
      }
    }
  ],
  "meta": { "page": 1, "page_size": 100, "total": 179, "total_pages": 2 }
}

• pricing.prompt — USD per input token. pricing.completion — USD per output token.
• Also: discount (fraction 0–1), request, web_search, input_cache_read, input_cache_write, internal_reasoning, and image/audio fields when applicable.
• Empty data is valid (e.g. no public pricing for that model_id on find/cheapest).
• Per-million display: multiply prompt / completion by 1_000_000.
• For POST /v1/calculate, pass provider_id from the row as the JSON field provider (e.g. "openai").

Reference providers — { "data": [ { "provider_id": "openai", "name": "OpenAI" }, … ] }

Reference models — { "data": [ { "provider_id": "openai", "model_id": "gpt-4o" }, … ], "meta": { … } }

---

GET /v1/reference/providers

curl 'https://api.narev.ai/v1/reference/providers'

---

GET /v1/reference/providers/{provider_id}

curl 'https://api.narev.ai/v1/reference/providers/openai'

---

GET /v1/reference/models

curl -G 'https://api.narev.ai/v1/reference/models' \
  --data-urlencode 'provider_ids=openai,anthropic' \
  --data-urlencode 'page=1' \
  --data-urlencode 'page_size=100'

---

GET /v1/price/{provider_id}

curl -G 'https://api.narev.ai/v1/price/openai' \
  --data-urlencode 'model_id=gpt-4o'

---

GET /v1/price/search

curl -G 'https://api.narev.ai/v1/price/search' \
  --data-urlencode 'q=gpt-4' \
  --data-urlencode 'page=1' \
  --data-urlencode 'page_size=100'

Omit q or use a broad fragment to page through the catalog (see meta.total_pages).

---

GET /v1/find/cheapest/{model_id}

curl 'https://api.narev.ai/v1/find/cheapest/gpt-4o'

Use an exact model_id from reference/models or price/search. Results sort by pricing.prompt ascending.

---

POST /v1/calculate

Given modelId, provider, and token usage, returns an itemized USD breakdown. Replaces legacy POST /models/pricing/calculate.

curl -X POST 'https://api.narev.ai/v1/calculate' \
  -H 'Content-Type: application/json' \
  -d '{
    "modelId": "gpt-4o",
    "provider": "openai",
    "subprovider": "OpenAI",
    "usage": {
      "promptTokens": 1000,
      "completionTokens": 500,
      "cacheReadTokens": 0,
      "cacheWriteTokens": 0,
      "reasoningTokens": 0
    }
  }'

200: costBreakdown.total (report to the user), plus pricing (input, output, request, inputCacheRead, inputCacheWrite, internalReasoning, webSearch), echoed usage, modelId, provider, subprovider.

Errors: 400 — usage validation (all five usage integers required; use 0 if unused). 402 — enterprise-only (pricing: null, error set). 404 — unknown modelId + provider (+ subprovider).

If calculate fails for a multi-host model, use GET /v1/price/search?q=… or GET /v1/find/cheapest/{model_id} to confirm provider_id, then retry with matching provider and subprovider when required.

---

When to use this skill

• Provider/model discovery, live token rates, cheapest host, or one-call USD.
• HTTP params, response fields, pagination, and calculate errors.

Snapshots in-repo → update-llm-pricing. Runtime billing in app → SDK / Next.js skills.

Constraints

• Only the seven endpoints above — no legacy list or calculate paths (/models/pricing/calculate, /v1/models/pricing/calculate, etc.).
• Calculate — POST /v1/calculate only.
• Do not hardcode rates; snapshot when you need pinned values.