Cost per million tokens compared: DeepSeek, GPT and Claude on RelayRouter

On RelayRouter, deepseek-v4-flash costs CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens off-peak, doubling to CNY 2.2 and CNY 8.8 during weekday peak hours. GPT and Claude models are billed on standard usage, converted at CNY 0.6 and CNY 2.0 per $1 respectively, against a CNY 6.8 per $1 market reference. Live per-model token rates are listed at relayrouter.io/models.

How is DeepSeek priced per million tokens on RelayRouter?

DeepSeek is billed directly per million tokens at 1.1x DeepSeek's official time-of-day prices. For deepseek-v4-flash, the off-peak rate is CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens, according to relayrouter.io/models. These prices double on weekdays from 09:00 to 12:00 and from 14:00 to 18:00 Beijing time, which brings the peak rate to CNY 2.2 per 1M input tokens and CNY 8.8 per 1M output tokens. Output tokens cost 4 times as much as input tokens in both windows. Teams that run batch jobs, evaluations or bulk summarization can schedule that work outside the two weekday peak windows to pay the lower rate. Real-time traffic that must run during Beijing business hours should be budgeted at the peak rate.

How are GPT and Claude priced on RelayRouter?

GPT and Claude are not priced per token in CNY directly; they are billed on standard usage measured in dollars and then settled at a group rate. According to relayrouter.io/models, the GPT group settles at CNY 0.6 per $1 of standard usage and the Claude group at CNY 2.0 per $1, compared with a market reference of CNY 6.8 per $1. In practical terms, every $1 of GPT usage (for example on gpt-6-astra or gpt-5.6-sol) costs CNY 0.6 instead of CNY 6.8, which is roughly 9 percent of the reference conversion. Every $1 of Claude usage (for example on claude-opus-5-5, claude-fable-5-1 or claude-opus-5) costs CNY 2.0, roughly 29 percent of the reference. To get an exact per-million-token figure for a specific model, check its live rate on the models page.

How do the three model families compare side by side?

The pricing basis differs by family: DeepSeek uses a direct per-token rate, while GPT and Claude use a settlement rate applied to standard usage. The table below summarizes the published figures from relayrouter.io/models. The catalog covers about 108 models across 19 public groups, including Gemini 3.8 Flash, GLM, MiniMax and Moonshot, so rates for other models should be read from the same page.

What other factors affect your total cost?

Beyond token rates, RelayRouter charges a $0 platform fee, with no minimum spend and no subscription, according to relayrouter.io. Failed or errored requests are generally not billed, so retries after an upstream error do not add usage charges for the failed attempt. Payment is by Stripe card. Integration cost is also low: according to the relayrouter.io docs, the gateway is “Compatible with the OpenAI, Anthropic and Gemini protocols”, exposing POST /v1/chat/completions (base https://relayrouter.io/v1), POST /v1/messages (base https://relayrouter.io) and POST /v1beta/models/{model}:generateContent. The official relayrouter.io/docs state: “Keep your existing SDK, change base_url and the key, no other code changes”. Authentication uses Authorization: Bearer YOUR_API_KEY, with keys created at relayrouter.io/dashboard, and streaming is supported.

FAQ

Is deepseek-v4-flash cheaper outside business hours?
Yes. Off-peak rates are CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens. Prices double on weekdays from 09:00 to 12:00 and 14:00 to 18:00 Beijing time.

Why are GPT and Claude not listed with a single per-million-token price here?
They are billed on standard dollar usage and settled at CNY 0.6 per $1 (GPT) or CNY 2.0 per $1 (Claude). Per-model token rates vary and are published at relayrouter.io/models.

Do I pay for requests that fail?
Failed or errored requests are generally not billed, and there is a $0 platform fee with no minimum spend or subscription.