llerandi/llm-price-tracker：每日更新的 LLM 价格追踪项目

A daily-updated reference of LLM model pricing across all major providers. One source of truth for input/output token costs, context windows, and capabilities - structured as JSON so you can consume it programmatically.

Live site: llerandi.github.io/llm-price-tracker - sortable, filterable table updated daily.

Pricing Table

Prices in USD per 1 million tokens. Sorted by provider, then by input price.

Batch and Cache Pricing

Some providers offer discounted rates for asynchronous (batch) processing and prompt caching. Prices in USD per 1 million tokens.

• Batch: requests are queued and processed asynchronously (typically within 24 hours) at ~50% off standard rates.
• Cache read: tokens served from the prompt cache at a fraction of the standard input cost.
• Cache write: tokens written to the cache, billed once at a slight premium over the standard input cost (Anthropic and GPT-5.6 models).

Batch and cache pricing is available for select models. See full pricing table →

Official batch and caching docs: Anthropic batch - Anthropic caching - OpenAI batch - OpenAI caching

Notes on specific models:

• Gemini 2.5 Pro: input $2.50/1M and output $15.00/1M above 200K tokens.
• GPT-5.6 Sol, Terra, and Luna: requests above 272K tokens charged at 2x input and 1.5x output.
• Perplexity Sonar models: token prices above exclude a per-request fee of $5-14/1K requests (varies by search context size).

Official pricing pages: AI21 Labs - Amazon Bedrock - Anthropic - Cohere - Fireworks AI - Google - Mistral - OpenAI - Perplexity - Together AI - xAI

Embeddable Badges

Add a live price badge to any README. Badges update daily and are served via jsDelivr CDN.

URL pattern:

https://img.shields.io/endpoint?url=https%3A%2F%2Fcdn.jsdelivr.net%2Fgh%2Fllerandi%2Fllm-price-tracker%40main%2Fdata%2Fbadges%2F{model-id}-{input|output|context}.json

Model IDs with slashes (e.g. Fireworks AI, Together AI) have the / replaced with - in the filename. Browse all IDs in data/prices.json or check data/badges/.

Example - Claude Sonnet 5 input price:

[![Claude Sonnet 5 input](https://img.shields.io/endpoint?url=https%3A%2F%2Fcdn.jsdelivr.net%2Fgh%2Fllerandi%2Fllm-price-tracker%40main%2Fdata%2Fbadges%2Fclaude-sonnet-5-input.json)](https://llerandi.github.io/llm-price-tracker/)

Use the live site to browse all models and copy badge embed code directly.

API Reference

All endpoints are static JSON files served via jsDelivr CDN with full CORS support (Access-Control-Allow-Origin: *). No API key required. Updated and CDN-purged daily at 06:00 UTC.

Base URL: https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main

Model IDs that contain / (Fireworks AI, Together AI) use - in badge filenames.

Schema stability

The JSON schema is stable. New fields may be added in future but existing fields will not be renamed or removed without a deprecation period. input_per_1m_usd, output_per_1m_usd, context_window_k, and all boolean capability fields are guaranteed to remain in the schema. Optional fields (batch_*, cache_*, notes) are present only when data is available; treat their absence or null as equivalent.

Endpoints

Get all models (curl):

curl https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/prices.json

Get models for a single provider (curl):

curl https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/providers/anthropic.json

Filter by price (Python, no dependencies):

import urllib.request, json

url = "https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/prices.json"
with urllib.request.urlopen(url) as r:
    data = json.load(r)

cheap = [m for m in data["models"] if (m["input_per_1m_usd"] or 999) < 1.0]
for m in cheap:
    print(f"{m['provider']} {m['model_name']}: ${m['input_per_1m_usd']}/1M in")

Filter by capability (JavaScript):

const res = await fetch(
  "https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/prices.json"
);
const { models } = await res.json();

// Models with vision support under $5/1M input
const visionModels = models.filter(
  m => m.supports_vision && (m.input_per_1m_usd ?? Infinity) < 5
);

Get a specific provider (JavaScript):

const res = await fetch(
  "https://cdn.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/providers/anthropic.json"
);
const { models } = await res.json();

JSON Schema

Each entry in models contains:

Top-level fields in prices.json:

Client Libraries

Installable wrappers around the jsDelivr JSON API. Both are zero-dependency and read-only.

JavaScript / TypeScript (Node >= 18 and browsers)

npm install llm-price-tracker

const { fetchPrices, getModel, getProvider } = require("llm-price-tracker");

// All models
const { models } = await fetchPrices();

// Single model
const sonnet = await getModel("claude-sonnet-5");
console.log(sonnet.input_per_1m_usd); // 2.00

// All models for a provider
const { models: anthropicModels } = await getProvider("anthropic");

Source: packages/npm/

Python (>= 3.9, no dependencies)

pip install llm-price-tracker

from llm_price_tracker import fetch_prices, get_model, get_provider

# All models
data = fetch_prices()

# Single model
sonnet = get_model("claude-sonnet-5")
print(sonnet["input_per_1m_usd"])  # 2.0

# All models for a provider
anthropic = get_provider("anthropic")

Source: packages/python/

Contributing

Prices change frequently. If you spot an outdated entry or a missing model, contributions are welcome.

1. Fork the repository and create a branch.
2. Edit data/prices.json with the correct values.
3. Run python scripts/validate.py to check the schema.
4. Run python scripts/verify_prices.py to check consistency (prices positive, batch <= standard, no duplicates).
5. Run python scripts/update_prices.py to regenerate the README table.
6. Open a pull request with a link to the official pricing page as evidence.

To add a new provider, add a new entry in data/prices.json following the existing schema.

How It Works

A GitHub Actions workflow (update.yaml) runs daily at 06:00 UTC. It runs scripts/update_prices.py, which:

1. Sorts all entries by provider and input price.
2. Updates the last_updated timestamp in prices.json.
3. Regenerates the pricing table in this README using HTML comment markers as boundaries.
4. Writes a compact daily snapshot to data/history/YYYY-MM-DD.json with just the pricing fields.
5. Commits and pushes if there are changes.

A second workflow (ci.yaml) runs on every push and pull request to lint the scripts and validate the JSON schema.

Troubleshooting

The live site or badges show stale data.

The daily workflow automatically purges the jsDelivr CDN cache after each push, so data is normally fresh within seconds. If you are still seeing stale data, hard-refresh the live site (Ctrl+Shift+R on Windows/Linux, Cmd+Shift+R on Mac). To force a manual purge for a specific file, open this URL in your browser:

https://purge.jsdelivr.net/gh/llerandi/llm-price-tracker@main/data/prices.json

Roadmap

Phase 1 - Foundation

• JSON schema definition
• Initial pricing data: Anthropic, Google, Mistral, OpenAI
• Auto-generated README pricing table from JSON
• CI pipeline (lint + JSON validation)
• Daily automated update workflow

Phase 2 - Expand

• Add providers: Cohere, Together AI, Fireworks AI, AI21 Labs
• Add batch pricing and prompt caching columns
• Track price history with daily snapshots

Phase 3 - Tooling

• GitHub Pages site with sortable and filterable table
• Embeddable price badge for other repositories
• REST-like endpoint via jsDelivr CDN with CORS support

Phase 4 - Insights

• Price history chart on the live site (visualize changes over time per model)
• Auto-generated changelog: markdown summary of price changes by date
• Cost calculator on the live site (tokens x price = estimated cost)
• Add providers: xAI (Grok), Perplexity, AWS Bedrock

Phase 5 - Community

• Automated price verification: script that cross-checks prices against official pages
• Support for multiple currencies (EUR, GBP) on the live site
• npm / PyPI package wrapping the jsDelivr JSON endpoint

Phase 6 - Automation and Discovery

• Automated GitHub Issue when a price changes by more than 10% (subscribe via Watch -> Issues)
• RSS/Atom feed of price changes, generated daily and served as static XML via jsDelivr CDN
• Add providers: DeepSeek, Groq, Qwen (Alibaba Cloud)
• Filters on the live site by tier, capabilities, and max price
• Model comparison view: select two models for a side-by-side diff

Phase 7 - Polish and Reach

• Dark mode with manual toggle (light by default, persisted in localStorage)
• data/latest.json permanent alias for the current prices
• Weekly price summary posted to GitHub Discussions every Monday
• SEO: sitemap.xml, Open Graph meta tags, JSON-LD structured data
• Embeddable context window badge (alongside existing price badges)
• Export current filtered table as CSV
• Shareable URLs: query params preserve active filters and comparison
• Changelog section on the live site
• Cost calculator presets (10-page PDF, code review, 1K support chats, article summary)
• Embedding model pricing (OpenAI, Google, Cohere, Voyage AI)
• More currencies: JPY, CAD, AUD (with fallback rates if API fails)
• Provider stats: aggregate summary per provider on the live site

Phase 8 - CDN Reliability and API

• Auto-purge jsDelivr CDN cache after every daily update (prices, badges, feed, providers)
• API documentation: stable schema contract, curl and JS/Python examples, versioning policy

Phase 9 - Coverage

• Add Azure OpenAI pricing
• Add Cerebras pricing

License

MIT