API documentation

Pricing Rules

Token pricing, cache accounting, image charges, and long-context multipliers.

AIGate calculates cost per request. Token models usually bill input, output, cache read, and cache write separately. Image models can use fixed or dynamic per-request pricing.

Token fields

Field	Meaning
input tokens	Normal text input tokens after separate cache/image/audio parts are accounted for.
output tokens	Generated text tokens.
cache read	Cached prompt tokens read from provider cache.
cache write	Tokens written into provider cache. Claude-style 5 minute and 1 hour writes can have different rates.
image input/output	Image tokens or image-generation call charges when a model reports them.

Long-context multipliers

Thresholds use total input context

The threshold is based on full input context length before cache discount. Cached tokens still count toward the threshold.

Provider/model family	Threshold	Input	Output	Cache
OpenAI models	272,000 tokens and above	x2	x1.5	x2
Google Pro models	200,000 tokens and above	x2	x1.5	x2
Grok models	Above 200,000 tokens	x2	x2	x2

For example, if a Google Pro request has 210K total input context, its input and cache portions use the long-context input price, and its output uses the long-context output price.

Cache hit percent

Cache hit percent is cache read divided by normal input plus cache read. Cache write is shown separately because writing cache is not a hit.

txt

cache_hit_percent = cache_read_tokens / (input_tokens + cache_read_tokens) * 100

Models

Balance