The rate card
What each model costs, to the million tokens
Every model BestRelay relays, with all four prices. There is no subscription and no minimum: you top up, and each call draws down by exactly what it used.
Anthropic
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| Claude Fable 5 | claude-fable-5 | Anthropic | $4.17 | $20.97 | $5.14 | $0.3472 |
| Claude Fable 5.1 | claude-fable-5-1 | Anthropic | $4.10 | $20.42 | $5.00 | $0.3472 |
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 | Anthropic | $0.4167 | $2.01 | $0.4861 | $0.0333 |
| Claude Opus 5 | claude-opus-5 | Anthropic | $2.29 | $11.46 | $2.85 | $0.1389 |
| Claude Opus 5.5 | claude-opus-5-5 | Anthropic | $1.74 | $8.13 | $2.01 | $0.0778 |
| Claude Sonnet 5 | claude-sonnet-5 | Anthropic | $1.32 | $6.67 | $1.67 | $0.1167 |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | Anthropic | $1.39 | $7.01 | $1.74 | $0.1222 |
OpenAI
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| GPT-5.6 Sol | gpt-5.6-sol | OpenAI | $0.4167 | $2.29 | $0.00 | $0.0333 |
| GPT-6 Astra | gpt-6-astra | OpenAI | $0.7639 | $3.96 | $0.00 | $0.0792 |
| GPT-6.1 Sol | gpt-6.1-sol | OpenAI | $0.1389 | $0.8333 | $0.00 | $0.0319 |
Google
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| Gemini 2.5 Flash | gemini-2.5-flash | Google | $0.2778 | $2.08 | $0.2778 | $0.0264 |
| Gemini 3.1 Pro | gemini-3.1-pro | Google | $1.74 | $9.93 | $1.74 | $0.1389 |
| Gemini 3.8 Flash | gemini-3.8-flash | Google | $1.25 | $5.56 | $1.25 | $0.1236 |
| Gemini 4 Argon | gemini-4-argon | Google | $1.67 | $8.06 | $1.67 | $0.0847 |
DeepSeek
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| DeepSeek V4 Pro | deepseek-v4-pro | DeepSeek | $1.53 | $4.58 | $0.00 | $0.0514 |
Qwen, from Alibaba
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| Qwen3.8 Max | qwen3.8-max | Qwen, from Alibaba | $0.8333 | $2.57 | $0.00 | $0.1014 |
Kimi, from Moonshot
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| Kimi K3 | kimi-k3 | Kimi, from Moonshot | $1.39 | $7.01 | $0.00 | $0.1389 |
GLM, from Zhipu
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| GLM-5.3 | glm-5.3 | GLM, from Zhipu | $0.625 | $1.94 | $0.00 | $0.1389 |
Grok, from xAI
| Model | Identifier to send | Provenance (maker) | Input | Output | Stored into cache | Served from cache |
|---|
| Grok 4.6 | grok-4.6 | Grok, from xAI | $0.625 | $1.81 | $0.00 | $0.1389 |
| Grok 4.7 | grok-4.7 | Grok, from xAI | $0.625 | $1.94 | $0.00 | $0.1389 |
Notes on this rate card
Each model has four prices. Input and output are the ordinary prompt and reply tokens; cache write applies when a provider stores a prompt prefix for reuse, and cache read applies when a later request hits that stored prefix. Prices of a yuan or more are set to the half yuan, smaller ones to the fen.
Every account is billed in yuan. Pages in other languages show an approximate dollar figure for reading convenience; the charge itself is computed in yuan from the tokens the provider reports.
Prices in US dollars per million tokens, converted at ¥7.2 to $1. Accounts are billed in yuan.
