model-pricing-calculator

This skill should be used when the user needs to fetch AI model pricing data from multiple API platforms, calculate model ratios, completion ratios, and group ratios based on a unified pricing formula, and output the results in a standardized JSON format.

A skill for fetching AI model pricing data from configured platform APIs and computing standardized pricing ratios (model ratio, completion ratio, group ratio).

When to Use

• Fetching model pricing information from AI API aggregation platforms
• Computing model ratios, completion ratios, and group ratios from raw pricing data
• Unifying pricing across multiple platforms using a consistent model ratio + group ratio scheme
• Reverse-calculating ratio configurations from target input/output prices
• Adding new pricing source URLs to the configuration

Core Concepts

Pricing Formula

text

Base price: $0.002 / 1K tokens (1 USD = 500,000 quota)

Mode 1 - Ratio-based (per-token):
  Cost = (InputTokens + OutputTokens × CompletionRatio) × ModelRatio × GroupRatio

Mode 2 - Fixed price (per-call, higher priority):
  Cost = ModelPrice(USD) × 500,000 × GroupRatio

Reverse Calculation

text

ModelRatio = InputPrice_per_1K / 0.002
CompletionRatio = OutputPrice_per_1K / InputPrice_per_1K

Workflow

Step 1: Check and Update URL Configuration

Read references/pricing_urls.json to verify configured platform URLs. To add a new platform, append a new entry with name, pricing_page, and api_endpoint fields.

The API endpoints typically follow the pattern {base_url}/api/pricing for platforms built on New API / One API systems.

Step 2: Fetch and Calculate Ratios

Run the calculation script:

bash

python scripts/fetch_and_calculate.py

Options:

• --urls-file <path> — custom URL config file path
• --output-dir <path> — save results as separate JSON files to the specified directory
• --verify — print a price verification table showing actual prices per group
• --no-snapshot — skip snapshot saving and diff comparison
• --models <patterns> — filter by model names, comma-separated, supports wildcard * (e.g. gpt-4*,claude-*)
• --groups <patterns> — filter by group names, comma-separated, supports wildcard * (e.g. default,aws*)
• --source <patterns> — filter by data source name, comma-separated, supports wildcard * (e.g. PackyAPI or 12AI,Packy*). Names correspond to name field in pricing_urls.json

The script will:

1. Fetch pricing data from each configured API endpoint
2. Extract model ratios, completion ratios, and group ratios
3. Unify duplicate models across platforms (first-source wins for ratios)
4. Apply --models / --groups filters if specified (supports wildcard *)
5. Output results in three standardized JSON blocks
6. Compare with previous snapshot (if exists, skipped in filter mode) and report any additions, removals, or value changes
7. Save current data as data/latest_snapshot.json (skipped in filter mode to avoid partial data overwriting full snapshot)

Filter Examples

Query specific models only:

bash

python scripts/fetch_and_calculate.py --models "gpt-4*,claude-opus*"

Query specific groups only:

bash

python scripts/fetch_and_calculate.py --groups "default,aws"

Combine both filters:

bash

python scripts/fetch_and_calculate.py --models "gpt-4o*" --groups "default,gemini*"

Query from a specific data source only:

bash

python scripts/fetch_and_calculate.py --source "PackyAPI"

Combine source, model and group filters:

bash

python scripts/fetch_and_calculate.py --source "PackyAPI" --models "claude-*" --groups "aws*"

Note: When --models, --groups, or --source is used, snapshot saving and diff comparison are automatically skipped to prevent partial data from overwriting the complete snapshot.

Step 3: Review Output Format

The output strictly follows this exact structure with 3 JSON blocks:

text

（1）模型倍率
{
  "claude-haiku-4-5-20251001": 1,
  "claude-opus-4-5-20251101": 2.5,
  "claude-opus-4-6": 2.5
}
（2）模型补全倍率
{
  "claude-haiku-4-5-20251001": 5,
  "claude-opus-4-5-20251101": 5,
  "claude-opus-4-6": 5
}
（3）分组倍率
{
  "default": 1.2,
  "aws": 2,
  "gemini": 0.6,
  "gemini-1": 3
}

Step 4: Snapshot and Diff

Each run automatically:

• Loads the previous snapshot from data/latest_snapshot.json (if it exists)
• Compares current results with the previous snapshot
• Reports all differences: added models, removed models, and changed values
• Overwrites the snapshot with the latest data

The diff report format:

text

============================================================
与上次数据对比（上次保存时间: 2025-03-23 10:00:00）
============================================================
  模型倍率:
    【新增 2 项】
      + new-model-a: 1.5
      + new-model-b: 2.0
    【删除 1 项】
      - old-model-x: 0.8
    【数值变化 1 项】
      * gpt-5: 0.625 → 0.75
  模型补全倍率: 无变化
  分组倍率: 无变化

Use --no-snapshot to skip this behavior.

Step 5: Cross-Platform Unification Strategy

When the same model appears on multiple platforms:

• Model ratio and completion ratio remain identical across platforms
• Price differences are controlled through group ratios assigned to each platform/channel
• Each platform's pricing channel corresponds to a specific group in the group_ratio map

For detailed pricing rules and formulas, refer to references/pricing_rules.md.

Bundled Resources

Resource	Purpose
scripts/fetch_and_calculate.py	Main script for data fetching, ratio calculation, snapshot and diff
references/pricing_urls.json	Platform URL configuration (add new sources here)
references/pricing_rules.md	Detailed pricing calculation rules and output format spec
data/latest_snapshot.json	Auto-generated snapshot of the latest run (created after first run)