Just the tokens.
Nothing in the way.
Prepaid usage. No recurring plan.
Input and output are priced separately, so you know what a request costs.
Below OpenRouter reference prices.
Qwen3.8 Flash, MiMo V2.6 Flash, MiMo V2.6 Pro, DeepSeek V4.1 Flash and GLM 5.3 Flash are priced 10% below their eligible OpenRouter reference prices.
| Model | Provider | Context window | Max output | Modality |
|---|---|---|---|---|
| Qwen3.8 Flash | Qwen | 1M | 131K | Text |
| MiMo V2.6 Flash | Xiaomi | 1M | 131.1K | Text |
| MiMo V2.6 Pro | Xiaomi | 1M | 131.1K | Text |
| DeepSeek V4.1 Flash | DeepSeek | 1M | 384K | Text |
| GLM 5.3 Flash | Z.ai | 1M | 131.1K | Text |
| Claude Fable 5.1 | Anthropic | 1M | 128K | Text |
| Claude Fable 5 | Anthropic | 1M | 128K | Text |
| Claude Haiku 4.5 | Anthropic | 8,192 (platform limit) | 1,024 | Text |
| Claude Opus 4.5 | Anthropic | 8,192 (platform limit) | 1,024 | Text |
| Claude Opus 4.6 | Anthropic | 8,192 (platform limit) | 1,024 | Text |
| Claude Opus 4.7 | Anthropic | 1M | 128K | Text |
| Claude Opus 4.8 | Anthropic | 1M | 128K | Text |
| Claude Opus 5.5 | Anthropic | 1M | 128K | Text |
| Claude Opus 5 | Anthropic | 1M | 128K | Text |
| Claude Sonnet 4.5 | Anthropic | 8,192 (platform limit) | 1,024 | Text |
| Claude Sonnet 4.6 | Anthropic | 1M | 128K | Text |
| Claude Sonnet 5 | Anthropic | 1M | 128K | Text |
| DeepSeek-V4-Flash-Vision-Exp | DeepSeek | 1M | 384K | Text |
| DeepSeek-V4-Flash | DeepSeek | 1M | 384K | Text |
| DeepSeek V4 Pro | DeepSeek | 1M | 384K | Text |
| Gemini 3.5 Flash-Lite | 1M | 65.5K | Text | |
| Gemini 3.5 Flash | 1M | 65.5K | Text | |
| Gemini 3.6 Flash | 1M | 65.5K | Text | |
| Gemini 3.8 Flash | 1M | 65.5K | Text | |
| GLM-5.1 | Z.AI | 200K | 128K | Text |
| GLM-5.2 | Z.AI | 1M | 128K | Text |
| GLM-5.3-FlashX | Z.AI | 1M | 131.1K | Text |
| GLM-5.3 | Z.AI | 8,192 (platform limit) | 1,024 | Text |
| GPT-5.6 Luna | OpenAI | 1.1M | 128K | Text |
| GPT-5.6 Sol | OpenAI | 1.1M | 128K | Text |
| GPT-5.6 Terra | OpenAI | 1.1M | 128K | Text |
| GPT-6 Astra | OpenAI | 1.1M | 128K | Text |
| GPT-6 Luna | OpenAI | 1.1M | 128K | Text |
| GPT-6 Sol | OpenAI | 1.1M | 128K | Text |
| Grok 4.5 | xAI | 500K (platform limit) | 1,024 | Text |
| Grok 4.6 | xAI | 500K (platform limit) | 1,024 | Text |
| Hy3 | Tencent | 256K | 128K | Text |
| Hy4 Preview | Tencent | 1M | 64K | Text |
| Kimi K2.6 | Moonshot AI | 8,192 (platform limit) | 1,024 | Text |
| Kimi K2.8 Preview | Moonshot AI | 1M (platform limit) | 1,024 | Text |
| Kimi K3 | Moonshot AI | 1M | 1M | Text |
| MiMo-V2.5-Pro | Xiaomi | 1M | 128K | Text |
| MiMo-V2.5 | Xiaomi | 1M | 128K | Text |
| MiniMax M2.7 | MiniMax | 204.8K | 131.1K | Text |
| MiniMax M3 | MiniMax | 1M (platform limit) | 1,024 | Text |
| Muse Spark 1.3 | Meta | 1M | 131K | Text |
| Qwen3.6-27B | Qwen | 128K | 32.8K | Text |
| Qwen3.7-Max | Qwen | 1M | 65.5K | Text |
| Qwen3.8-27B | Qwen | 262.1K | 131.1K | Text |
| Qwen3.8-Max | Qwen | 1M | 131.1K | Text |
| DeepSeek V3.2 | DeepSeek | 164K | 164K | Text |
| Gemini 3.1 Pro | 1M (platform limit) | 64K | Text | |
| Gemini 3 Flash | 8,192 (platform limit) | 1,024 | Text | |
| GPT-5.4 | OpenAI | 272K | 128K | Text |
| GPT-5.5 | OpenAI | 1.1M | 128K | Text |
| GPT-5.5 Instant | OpenAI | 1M | 128K | Text |
| GPT-5.4 Pro | OpenAI | 1.1M | 128K | Text |
| GPT-5.2 | OpenAI | 8,192 (platform limit) | 1,024 | Text |
| GPT-5.4 Mini | OpenAI | 400K | 16K | Text |
| GPT-5 Mini | OpenAI | 8,192 (platform limit) | 1,024 | Text |
| GPT-5.4 Nano | OpenAI | 400K (platform limit) | 1,024 | Text |
| GPT-5 Nano | OpenAI | 8,192 (platform limit) | 1,024 | Text |
| Kimi K2.5 | Moonshot AI | 256K | 256K | Text |
| JEV-1.13.0 | TypeSafe AI | 32K | Structured result | Structured decisions |
| GPT-Image-2 | OpenAI | 1024×1024 image | One PNG | Image generation |
Live per-token prices load from the catalogue API and refresh hourly. This table lists stable model facts only.
How pricing works
Cache write entries marked “Input rate” show the cost of uncached input tokens for automatic caching; no separate cache-write rate is quoted.
Standard prices include separate cache rates where supported. New DeepSeek models follow the official peak and off-peak schedule. Image generation is billed by tokens, not by image. JEV decisions charge input tokens only.
For the five existing discounted models, prices are checked against synchronous endpoints for the same model. Scheduled prices are applied at the time of verification. Input and output minima may come from different endpoints. Batch tariffs are excluded.
Reference endpoints can use different quantization and context limits. This is a price comparison, not a claim of identical performance. Expired prices cannot authorize a request.
Explore the reference catalogueStay in control
Each request reserves its maximum possible cost, then settles against the tokens actually reported. Uncertain requests keep their reservation until reviewed.
Credit purchases are in USD.
Create an account