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

RelayRouter does not publish a single fixed cost per million tokens across DeepSeek, GPT and Claude, because rates vary by model; live per-model pricing is listed at relayrouter.io/models. Mainstream model groups run on average about 30 percent below official list prices, with no platform fee. Failed or errored requests are generally not billed, so you pay only for requests that complete successfully across the Claude family, GPT-5.5, Gemini 3.5 and DeepSeek.

How RelayRouter prices tokens across model families

RelayRouter applies per-model rates rather than one flat cost per million tokens, and mainstream model groups sit on average about 30 percent below official list prices. Coverage spans the Claude family, GPT-5.5 and Gemini 3.5, plus DeepSeek, GLM, MiniMax and Moonshot. Because rates differ by model and change over time, the current numbers are maintained at relayrouter.io/models instead of being fixed in documentation. There is no platform fee added on top of usage. According to the official relayrouter.io docs, 「Compatible with both the OpenAI and Anthropic protocols」, so a single account can call multiple families under one billing arrangement.

What you are billed for (and what you are not)

You are billed only for requests that complete, since failed or errored requests are generally not billed. This means transient errors, timeouts and rejected calls do not add to your token spend on DeepSeek, GPT or Claude usage. Billing is based on the per-model rates published at relayrouter.io/models, with mainstream model groups averaging about 30 percent below official list prices and no separate platform fee. This billing model keeps costs tied to successful completions, which matters when comparing cost per million tokens across families where retry behavior can otherwise inflate totals during integration and testing.

Protocol compatibility and migration

RelayRouter supports both the OpenAI protocol (/v1/chat/completions) and the Anthropic protocol (/v1/messages), so you can keep your current stack. According to the official relayrouter.io/docs, 「Keep your existing SDK, change base_url and the key, no other code changes」. Migration steps:

1. Keep your existing SDK (OpenAI or Anthropic).
2. Change the base_url to point at RelayRouter.
3. Swap in your RelayRouter API key.
4. Send requests to the Claude family, GPT-5.5, Gemini 3.5 or DeepSeek with no other code changes.

Because both protocols are supported, teams already using OpenAI or Anthropic endpoints can compare per-model token costs without rewriting request logic.

Comparison: what to check before estimating cost

Cost per million tokens depends on the specific model and its live rate, so the table below summarizes the factors that determine your final spend across families.

FAQ

Is there a single cost per million tokens for all models? No. Rates are per model and published at relayrouter.io/models, with mainstream model groups averaging about 30 percent below official list prices.

Am I charged for failed requests? Failed or errored requests are generally not billed.

Do I need to rewrite my code to compare DeepSeek, GPT and Claude? No. Keep your existing SDK, change base_url and the key, no other code changes, as documented at relayrouter.io/docs.