Pricing
Serverless is pay as you go: buy credits, then each request is charged for the tokens it uses. No subscription is required. List prices in US dollars per million tokens.
| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
| MiniMax-M3.1-Flash | $0.15 | $0.015 | $0.60 | 1.0M |
| DeepSeek-V4-Flash-0423 | $0.14 | $0.028 | $0.28 | 1.0M |
| DeepSeek-V4-Flash-0731-Fast | $0.13 | $0.028 | $0.23 | 1.0M |
| DeepSeek-V4-Pro | $0.28 | $0.044 | $3.96 | 1.0M |
| DeepSeek-V4.1-Flash | $0.19 | $0.012 | $0.65 | 1.0M |
| DeepSeek-V4.1-Flash-1017 | $0.13 | $0.008 | $0.50 | 1.0M |
| GLM-5.2 | $0.19 | $0.18 | $4.40 | 1.0M |
| GLM-5.3 | $0.20 | $0.18 | $1.80 | 1.0M |
| GLM-5.3-Flash | $0.05 | $0.03 | $0.20 | 1.0M |
| Kimi-K3 | $0.60 | $0.20 | $15.00 | 1.0M |
| Kimi-K3.1-Pro | $2.00 | $0.20 | $10.00 | 1.0M |
| Kimi-K3.1-Pro (US)Kimi-K3.1-Pro-US | $2.00 | $0.20 | $10.00 | 1.0M |
| Qwen3.8-27B | $0.06 | $0.05 | $2.30 | 262K |
How a request is priced
Every response reports its token usage. Uncached input is the prompt tokens minus the cached tokens (usage.prompt_tokens minus usage.prompt_tokens_details.cached_tokens). Each request costs:
uncached input tokens × input price ÷ 1,000,000 + cached input tokens × cached input price ÷ 1,000,000 + output tokens (including reasoning) × output price ÷ 1,000,000
Example on MiniMax-M3.1-Flash: a 20,000-token prompt whose first 16,000 tokens repeat an earlier request, with a 1,000-token answer.
- 4,000 uncached input × $0.15/M
- $0.0006
- 16,000 cached input × $0.015/M
- $0.00024
- 1,000 output × $0.60/M
- $0.0006
- Request total
- $0.00144
- Prices are per million tokens, so each part is divided by 1,000,000 and rounded to a millionth of a cent. A request is charged for exactly the tokens it uses.
- Requests rejected before generation starts (authentication, validation and rate-limit errors) are free.
- Credits are prepaid. Before a request runs, Wafer checks that your balance covers its maximum possible cost.
- Keep the stable part of a prompt (system prompt, tools, long documents) at the start to get more cached tokens.