One OpenAI-compatible API, priced in USD. Every rate matches the upstream provider's public price — we add zero markup and earn only the 10% top-up fee.
| Model | Provider | Context | Input USD / 1M tokens | Output USD / 1M tokens | Cache-hit input USD / 1M tokens | Upstream / data residency | Capabilities |
|---|---|---|---|---|---|---|---|
deepseek-flash | DeepSeek | 1M | $0.15 | $0.6 | $0.003 | DeepSeek first-party API (api.deepseek.com) | ChatCodeFunction callingVisionPeak/off-peak |
qwen-plus | Alibaba Qwen | 1M | $0.1096 | $0.274 | $0.011 | Alibaba Cloud Model Studio (Mainland China node) | BalancedLong contextThinking modeTiered pricing |
qwen-max | Alibaba Qwen | 32K | $0.3288 | $1.3151 | $0.0329 | Alibaba Cloud Model Studio (Mainland China node) | FlagshipReasoning |
qwen-flash | Alibaba Qwen | 1M | $0.0205 | $0.2055 | $0.0021 | Alibaba Cloud Model Studio (Mainland China node) | LightweightLow costChat |
auto | AIGateway | per routed model | Billed at whichever model it routes to — routing itself is free | Smart routing | |||
| deepseek-flash time-of-day rates | Input | Output | Cache-hit input |
|---|---|---|---|
| Off-peak (default) | $0.15 | $0.6 | $0.003 |
| Peak | $0.3 | $1.2 | $0.006 |
| qwen-plus context tiers | Input length per request | Input | Output |
|---|---|---|---|
| Standard | ≤ 128K | $0.1096 | $0.274 |
| Tier 2 | ≤ 256K | $0.3288 | $2.7397 |
| Tier 3 | ≤ 1000K | $0.6575 | $6.5753 |
Not sure which model to use? Send model: "auto" and we pick a tier based on request complexity. You are billed at the model actually used — routing is free.
| Complexity tier | Routes to | Input $/1M | Output $/1M | — |
|---|---|---|---|---|
SIMPLE | qwen-flash | $0.0205 | $0.2055 | Greeting, classification, extraction |
MEDIUM | qwen-plus | $0.1096 | $0.274 | General writing, summarization |
COMPLEX | deepseek-flash | $0.15 | $0.6 | Multi-step technical tasks, code |
REASONING | qwen-max | $0.3288 | $1.3151 | Proofs, deep analysis |
Price table version 2026-09-11.1. When a provider changes rates we follow and publish it in the changelog. Every invoice line is computed with the price table in effect at the moment of the request, so historical bills never shift.