Claude Code Relay Pricing and Paid API Evidence
Compare Claude Opus 5 input and output rates, source dates, paid request results, client restrictions, and best-fit scenarios across six AI relay providers.
Confirm client restrictions first
Claude Code compatibility does not guarantee that a general Anthropic SDK will work. Some keys are limited to Claude Code or VS Code, so the protocol, base URL, and required models must be tested in the intended client.
Compare the same model and route class
The evidence table fixes the model at Claude Opus 5 and keeps provider-published prices separate from our paid request results. A passing call is not attributed to a price group unless the tested key's route is known.
Who this is for
This guide is for developers who primarily use Claude Code, can test a small balance first, and accept possible pool quotas, concurrency rules, or client-only restrictions.
Claude Code relay evidence snapshot
The same Claude Opus 5 model is used for the comparison. Prices are provider-published premium-route rates; request results come from AI Relay Check test accounts and do not prove that the tested key used the quoted group.
| Provider | Provider group | Input / 1M | Output / 1M | Price date | Passed / total | Latest test | Best suited for |
|---|---|---|---|---|---|---|---|
| LinksAPI | Dedicated CC Max | ¥3.4 | ¥17 | 2026-09-25 | 4 / 4 | 2026-09-29 | Users prioritizing a provider-labeled premium route |
| H API | Claude Max | ¥6.4 | ¥32 | 2026-09-25 | 3 / 3 | 2026-09-25 | Claude Code or VS Code users who accept a client-only route |
| UU API | CC full-capability MAX - CC client only | ¥6.8 | ¥34 | 2026-09-25 | 1 / 3 | 2026-09-25 | Claude Code or VS Code users who accept a client-only route |
| APINebula | Claude Code global protected route | ¥6.8 | ¥34 | 2026-09-25 | 3 / 4 | 2026-09-29 | Claude Code or VS Code users who accept a client-only route |
| boxying | Claude official route | ¥8 | ¥40 | 2026-09-25 | 2 / 11 | 2026-09-25 | Users prioritizing a provider-labeled premium route |
| OpenOx | direct | ¥19.2 | ¥96 | 2026-09-24 | 3 / 3 | 2026-09-27 | Users prioritizing a provider-labeled premium route |
What a successful call proves
A completed request verifies the tested model ID, key, protocol and route at that moment. It does not prove model identity, every Claude model, long-term stability, or wallet billing. Client-only keys must be tested through the designated Claude Code flow.
Five checks before connecting Claude Code
Client and protocol
Confirm whether the key is restricted to Claude Code, VS Code, Anthropic Messages, or another designated client.
Base URL and authentication
Use the provider's current HTTPS endpoint and the authentication format documented for the tested route.
Exact model ID
Verify the requested Claude model with a generation call. A model-list response alone is not enough.
Streaming and tools
Test streaming, tool use, and context length separately when the workload depends on them.
Route and billing
Record the test key's group and reconcile usage or balance before attributing a published price to the request.
Frequently asked questions
How should I choose a Claude Code relay?
Start with the exact Claude model and intended client. Then compare the same route class by input and output price, paid request samples, client restrictions, concurrency, quota, support terms, and same-route billing evidence.
Does Claude Code compatibility mean the Anthropic API also works?
No. Some keys and groups are client-only. Anthropic Messages, streaming, tool use, and other SDK workflows require their own generation tests.
What is the difference between a Claude Code pool and an official or direct route?
A pool may share subscription capacity and impose quota, concurrency, or rotation rules. Official and direct are provider labels until separately verified, so the two route classes should not share one price ranking.
Is the cheapest Claude relay the best recommendation?
Not by itself. A lower rate may apply to a reverse, pooled, cached, subscription-only, or client-specific route. Compare price only after matching the model, route, protocol, and billing conditions.
Answer summary
- Latest test date
- Providers covered
- 6
- Models / protocols
- claude-fable-5 / claude-fable-5-1 / claude-opus-5 +1 more · Anthropic / ClaudeCode / OpenAI
- Successful / total samples
- 69 / 122
- Price / billing verified
- 6/6 price · 2/6 billing
- Still unverified
- Model identity, every Claude Code workflow, future uptime and unmatched billing
- One-sentence conclusion
- Compare the same Claude model and route class; a passing request does not prove every Claude Code workflow or the upstream model identity.