Model Pricing Calculator

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

CODEBLOCK0

Reverse Calculation

CODEBLOCK1

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:

CODEBLOCK2

Options:

• - --urls-file <path> — custom URL config file path
• INLINECODE6 — save results as separate JSON files to the specified directory
• INLINECODE7 — print a price verification table showing actual prices per group
• INLINECODE8 — skip snapshot saving and diff comparison
• INLINECODE9 — filter by model names, comma-separated, supports wildcard * (e.g. gpt-4*,claude-*)
• INLINECODE12 — filter by group names, comma-separated, supports wildcard * (e.g. default,aws*)
• INLINECODE15 — filter by data source name, comma-separated, supports wildcard * (e.g. PackyAPI or 12AI,Packy*). Names correspond to name field in INLINECODE20

The script will:

1. 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:
CODEBLOCK3

Query specific groups only:
CODEBLOCK4

Combine both filters:
CODEBLOCK5

Query from a specific data source only:
CODEBLOCK6

Combine source, model and group filters:
CODEBLOCK7

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:

CODEBLOCK8

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:

CODEBLOCK9

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