AI Model Pricing Data — 70+ Models, One JSON File

CC BY 4.0 · USD per 1M tokens · prices verified 2026-09-30

A single static JSON file with the input and output price per 1,000,000 tokens, context window and tokenizer for 70 models across 22 providers — OpenAI, Anthropic, Google, Meta, xAI, Mistral, Amazon, DeepSeek, Qwen, GLM, Kimi, Doubao, ERNIE, Hunyuan, Spark and more. Free to reuse in your app, dashboard, docs or article, no key and no rate limit. It is the same dataset that powers the AI Token Calculator.

Fetch it

import json, urllib.request
data = json.loads(urllib.request.urlopen(
    "https://tokenestimate.com/data/model-prices.json").read())
for m in data["models"]:
    print(m["label"], m["priceInputPer1M"], m["priceOutputPer1M"])

What each record contains

Sample

{
  "_meta": {
    "name": "AI model token pricing table",
    "endpoint": "https://tokenestimate.com/data/model-prices.json",
    "license": "CC BY 4.0",
    "lastVerified": "2026-09-30",
    "currency": "USD",
    "unit": "price per 1,000,000 tokens",
    "citation": "TokenCalc, \"AI Model Token Pricing Data\", tokenestimate.com, 2026."
  },
  "models": [
    {
      "id": "gpt-5.5",
      "label": "GPT-5.5",
      "provider": "OpenAI",
      "tokenizer": "tiktoken:o200k_base",
      "contextWindow": "1M",
      "priceInputPer1M": 5,
      "priceOutputPer1M": 30
    },
    {
      "id": "claude-opus-5-5",
      "label": "Claude Opus 5.5",
      "provider": "Anthropic Claude",
      "tokenizer": "estimate",
      "estimate": { "tokensPerCJKChar": 1.2, "charsPerToken": 4 },
      "contextWindow": "1M",
      "priceInputPer1M": 4,
      "priceOutputPer1M": 20,
      "note": "flagship, from 2026-09-22"
    },
    {
      "id": "embedding-3-small",
      "label": "text-embedding-3-small",
      "provider": "OpenAI embedding",
      "tokenizer": "tiktoken:cl100k_base",
      "contextWindow": "–",
      "priceInputPer1M": 0.02,
      "priceOutputPer1M": null
    }
  ]
}

Coverage

License and citation

• CC BY 4.0 — copy it, transform it, sell a product built on it.
• Attribution: a link to https://tokenestimate.com/ or the citation line below, wherever you publish the data.
• Access-Control-Allow-Origin: * — no CORS workaround needed.
• It is a static file: cache-bust with a query string if you poll it, and re-check _meta.lastVerified rather than assuming freshness.

How the numbers are gathered

Each price is the provider's public list price in USD, read from the vendor's own pricing page and cross-checked against a second source where one exists. The verification date in _meta.lastVerified applies to the whole file. Not included: batch, cached-input, fine-tuned, provisioned-throughput or contract rates, and image/video/audio billing that some providers express per second or per image instead of per token. Prices move — treat this as a dated snapshot, not a live feed.

Pricing data FAQ

Is the AI model pricing JSON free to use?

Yes. It is licensed CC BY 4.0: use it commercially, modify it, redistribute it. The only requirement is attribution — a link to tokenestimate.com or the citation line TokenCalc, "AI Model Token Pricing Data", tokenestimate.com, 2026.

How many models and providers does the dataset cover?

70 models across 22 provider groups as of 2026-09-30, spanning OpenAI, Anthropic, Google, Meta, xAI, Mistral, Amazon and the major Chinese labs (DeepSeek, Qwen, GLM, Kimi, Doubao, ERNIE, Hunyuan, Spark, MiniMax, 01.AI, StepFun, SenseTime, Baichuan, Kunlun, Xiaomi, Huawei).

Are the prices exact?

They are the providers' published list prices, verified 2026-09-30, in USD per 1M tokens. Discounted tiers — batch, cached input, long-context surcharges, contract pricing — are not modelled, so check the provider page before you publish a number or send a bill.

How often is the dataset updated?

Whenever model pricing changes; the file carries its own _meta.lastVerified date, so you can decide for yourself whether a value is current instead of trusting a changelog.

Why do some records say "estimate" in the tokenizer field?

OpenAI publishes tiktoken, so 18 of the records are counted with the official BPE tokenizer. Other vendors do not publish their tokenizer, so token counts for them are calibrated character-ratio estimates (about 4 characters per English token), accurate to roughly 5–10%. The tokenizer field tells you which case a record is in.

Can I hotlink the JSON from my app?

Yes, and it is served with permissive CORS. The URL is stable, and because it is a plain static file it will keep working without a key, quota or plan.

What does a null priceOutputPer1M mean?

The model has no output price because it produces embeddings rather than generated text — e.g. text-embedding-3-small at $0.02 per 1M input tokens. Treat null as "not applicable", not as "free".