Pricing, metering & prompt caching

Per-token, per request

Billing is post-paid against a prepaid wallet: every request is metered on actual tokens used, and the response reports exactly what it cost.

cost = prompt_tokens_miss × prompt_rate
     + prompt_tokens_cached × input_cache_read_rate
     + completion_tokens × completion_rate

All rates are USD per 1M tokens, published live at GET /api/v1/models. Example (glm-5.3-flash): 1,000 miss input tokens + 50,000 cached input tokens + 2,000 output tokens:

0.001 × $0.15  = $0.00015
0.050 × $0.03  = $0.00150
0.002 × $0.50  = $0.00100
                = $0.00265 total

How billing executes

Before your request runs, an estimate (prompt + worst-case max_tokens) is reserved from your balance. When the stream completes, the reservation is settled against actual usage and the difference is refunded automatically. If the request fails upstream or you disconnect mid-stream, the reservation is refunded in full — you are only ever charged for delivered tokens.

Prompt caching (the discount you get for free)

Models with prompt caching bill repeated identical prefixes at the subsidized input_cache_read rate — up to 10× cheaper than a miss. The split is automatic and reported per request:

"usage": {
  "prompt_tokens": 51234,
  "prompt_tokens_details": { "cached_tokens": 50176 },
  "completion_tokens": 312,
  "cost": 0.0192
}

To maximize hits: keep your system prompt (schemas, few-shots, RAG corpus) static and byte-identical at the start of messages, and put variable content last. See the blog post "Your prompts are cached" for patterns.

Reading your spend

• usage.cost in every response — per-request truth
• GET /api/v1/key — live balance + in-flight reservations
• Console → Usage — per-model spend, latency percentiles, daily charts
• Console → Billing — full ledger (every charge, top-up, refund)