AI Pricing Data

Automatically maintained JSON database of Claude / Gemini / OpenAI model pricing, capabilities, and API endpoint information.

A GitHub Actions workflow runs every Monday at UTC 00:00 and commits any changes to pricing.json. Desktop clients can fetch the latest data with a single HTTP GET — no API key required.

---

Client fetch URL

https://raw.githubusercontent.com/EvanYu34/pricing-data/main/pricing.json

Example (curl):

curl -s https://raw.githubusercontent.com/EvanYu34/pricing-data/main/pricing.json \
  | python -m json.tool

The personal-stack family consumes this via ai_selector.default_catalog(merge_pricing=True) — the catalog overlays this JSON onto the canonical 5-vendor model list to drive the UI dropdowns.

---

Local development

Prerequisites

• Python 3.11+
• pip install -r requirements.txt

Run all scrapers

python main.py

Run a single provider

python main.py --provider claude
python main.py --provider gemini
python main.py --provider openai

Enable Playwright locally

By default, Playwright is only activated inside GitHub Actions. To use it locally (useful for debugging JS-heavy pages):

# Install Playwright browsers once
python -m playwright install chromium

# Run with Playwright enabled
USE_PLAYWRIGHT=1 python main.py

Output

• pricing.json — updated in place
• .commit_message — the git commit message that would be used by CI

---

JSON schema

{
  "last_updated": "2025-03-01T08:00:00Z",   // ISO-8601 UTC
  "schema_version": "2.0",
  "sources": {
    "<provider>": {
      "fetch_status": "success" | "failed",
      "error_message": null | "<reason>",
      "provider_info": { ... },
      "models": [ { ... } ]
    }
  }
}

Model object

Field	Type	Description
model_id	string	Canonical API identifier (e.g. claude-3-5-sonnet-20241022)
display_name	string	Human-readable name
context_window_tokens	int | null	Maximum input context in tokens
capabilities	string[]	See capability table below
pricing	object	See pricing fields below
api_endpoints	object[]	See endpoint types below
multilingual	bool	Whether the model supports non-English input/output
is_deprecated	bool	Whether the model is deprecated / retired
notes	string	Any additional notes
source	string	Where the current pricing values come from: litellm / scraper / merged / stale / fallback. Used by the audit to gate which fields it cross-validates.
last_seen_run	string (YYYY-MM-DD)	Last run date on which any source returned this model_id. Stale if it falls behind last_updated.
last_verified_at	string | null (ISO-8601 UTC)	Finer-grained timestamp of the last successful verification of this model_id this run. null for legacy entries that haven't been re-hit yet. Downstream consumers can compare to last_updated to detect stale model entries (e.g. price hasn't been re-verified for N weeks → use with care).
verified_source	string | null	The actual source that returned this model_id during its last verification: litellm / scraper / fallback. Distinct from source — that one describes where the pricing values come from (and can be merged / stale), while verified_source describes which scraper layer actually saw the model.

Pricing fields

Field	Unit	Description
currency	—	Always "USD"
input_per_1m_tokens	USD	Cost per 1 000 000 input tokens
output_per_1m_tokens	USD	Cost per 1 000 000 output tokens
cache_write_per_1m_tokens	USD	Prompt-cache write cost (Claude)
cache_read_per_1m_tokens	USD	Prompt-cache read cost (Claude)
per_minute_audio	USD	Audio transcription cost per minute (Whisper)
per_image	USD	Image generation cost per image (DALL-E)
audio_input_per_1m_tokens	USD	Audio input cost (GPT-4o Audio)
audio_output_per_1m_tokens	USD	Audio output cost (GPT-4o Audio)
notes	string	Supplementary pricing notes

---

Capability definitions

Value	Description
text_generation	General text generation, Q&A, writing
translation	Multi-language translation
transcription	Speech-to-text (STT) — e.g. Whisper
text_to_speech	Text-to-speech (TTS)
image_understanding	Image input, OCR, visual Q&A (multimodal)
image_generation	Image generation — e.g. DALL-E, Imagen
document_processing	Long-document understanding, PDF processing
code_generation	Code generation and debugging
embedding	Text vectorisation for retrieval / similarity
function_calling	Tool use / Function Calling support
structured_output	JSON mode or guaranteed structured output

---

API endpoint types

endpoint_type	HTTP path	Description
chat_completions	POST /v1/chat/completions	OpenAI-compatible chat format
messages	POST /v1/messages	Anthropic native format
generate_content	POST /v1beta/models/{model}:generateContent	Google native format
audio_transcriptions	POST /v1/audio/transcriptions	Audio → text (Whisper)
audio_speech	POST /v1/audio/speech	Text → audio (TTS)
embeddings	POST /v1/embeddings	Text → vector
images_generations	POST /v1/images/generations	Text → image

---

Project structure

pricing-data/
├── .github/
│   └── workflows/
│       └── update_pricing.yml     # GitHub Actions — weekly auto-update
├── scrapers/
│   ├── __init__.py
│   ├── base_scraper.py            # Abstract base with fetch helpers
│   ├── claude_scraper.py          # Anthropic scraper
│   ├── gemini_scraper.py          # Google Gemini scraper
│   ├── openai_scraper.py          # OpenAI scraper
│   └── capabilities_fallback.py  # Hard-coded capability / endpoint data
├── utils/
│   ├── __init__.py
│   └── json_merger.py             # Merge logic + diff summary
├── main.py                        # CLI entry point
├── pricing.json                   # Auto-maintained data file
├── requirements.txt
└── README.md

---

Adding a new provider

1. Create scrapers/<provider>_scraper.py extending BaseScraper.
2. Implement scrape_pricing(), scrape_models(), and build_provider_data().
3. Register it in main.py's SCRAPERS dict.
4. Add fallback capability / endpoint data to scrapers/capabilities_fallback.py.

---

Updating fallback data

Edit scrapers/capabilities_fallback.py to add or correct model capability lists, context window sizes, display names, and endpoint configurations. These values are used when live scraping fails or returns incomplete data.

---

License

MIT