Direct conclusion: You can't generalize about relay services — whether a service is "reliable" depends on whether it's willing to publicly disclose three things: its upstream channel, its rate table, and its billing rules. Services that disclose all three and allow line-by-line verification have a high probability of being trustworthy. Bargain services that only quote prices privately, are vague about their upstream, or use monthly cards to push you into囤积 quota carry extremely high risk. The method for evaluating them isn't about reading marketing copy — it's about five signals you can test yourself: model substitution, quota expiration, charging for failures, upstream origin, and rate transparency. This article breaks each signal into actionable self-test steps that work for Claude, CC (Claude Code), and Codex relay services alike.
Are Relay Services Reliable? A One-Sentence Criterion
The standard for judging whether a relay service is reliable comes down to one word: verifiability (transparency).
- Can it publicly disclose its upstream (official API / Bedrock)? Does it publish a pricing page with real-time rates? Does it only charge for successful requests? If all three are true, it's worth using long-term.
- Conversely, if the upstream is unclear, prices are only quoted privately, and monthly card quota expires unused — no matter how cheap it is, steer clear.
Low prices themselves aren't the problem — the source of the low prices is what matters. Being cheap through cache hits and volume procurement is legitimate; being cheap through model substitution, inflated rates, or account pool upstream is adulteration. This article teaches you how to tell the difference.
How to Tell if a Relay Service Is Adulterated: Five Self-Test Signals
Signal 1: Does It Substitute Models During Peak Hours (Quality Degradation)?
The most common trap: advertising Opus/Sonnet but actually routing to cheaper models during peak evening hours. Self-test method: run the same complex task during local peak hours (8 PM–12 AM) and compare output quality and response characteristics with daytime results. A more direct approach is to ask the model to identify itself and perform tasks that only the higher-tier model can reliably complete (long-context reasoning, complex refactoring), then compare with official API results.
Signal 2: Does Your Quota Expire?
Read the billing terms carefully: is it pay-as-you-go, or does it use "monthly card /套餐 quota that expires unused"? Under the monthly card model, the provider has an incentive to make sure you don't use your full quota. With pay-as-you-go and non-expiring balances, your interests are aligned. This is also a core indicator of whether the provider might shut down — the prepaid model inherently creates an incentive to "push囤积 quota then close up shop."
Signal 3: Are Failed Requests Charged?
Intentionally trigger a 500/429 error (e.g., by instantly maxing out concurrency) and check your bill. Legitimate gateways only charge for successfully completed requests. Services that also charge for failures will cost you far more than the listed price over time.
Signal 4: Is the Upstream Channel Disclosed?
Ask customer support directly: Is the upstream official API, AWS Bedrock, or "proprietary channel / stable supply"? Vague answers that only mention "stable supply" most likely indicate reverse-engineered APIs or shared account pools — where the ban risk is on you.
Signal 5: Is the Rate Table Transparent?
Require a publicly available, timestamped pricing page showing per-model input/output token rates, cache pricing, and discount rules. Services without a public pricing page that only quote prices privately should be eliminated immediately.
How to Verify If a Relay Service's Opus Is Genuine
This is one of the most frequently asked questions in the community. Three-step self-test:
- Self-identification + difficult task: Ask the model to complete a task that only the Opus tier can reliably handle (e.g., multi-step reasoning in an very long context, cross-file complex refactoring). See if it "gets dumber" mid-task.
- Compare with the official API: Send the same prompt through both the relay service and the official API, then compare output quality, style, and attention to detail.
- Check the response metadata: Legitimate gateways pass through the model identifier and usage (token count) unmodified. If tool calling, streaming, or cache control headers are stripped, the request was likely rewritten — which is itself a red flag.
Remember: The real thing can withstand comparison. Adulteration is immediately exposed when compared.
Will Using a Relay Service Get My Account Banned?
Ban risk comes almost entirely from the upstream form, not from "using a relay service" itself:
- Reverse-engineered APIs / shared account pools: If one upstream account gets suspended, all users are affected in a chain reaction — this is the primary source of bans.
- Gateways using official API channels: Every call you make is a legitimate API request. There's no account pool, so no chain reaction ban risk.
So "will I get banned" is essentially equivalent to Signal 4 — the upstream source. This is also the most tangible security advantage of direct gateways over traditional account pool relays and mirror sites.
What's Special About CC (Claude Code) and Codex Relay Services?
Agent tools like Claude Code and Codex generate highly repetitive requests (system prompts, tool definitions, history context appear repeatedly), so there are two additional things to check:
- Cache hit rate: Cached requests are billed at cache pricing, which is far cheaper than full-price tokens. A high cache hit rate is the legitimate path to "being reasonably cheap."
- Agent protocol compatibility: Can tool calling, streaming, and cache control headers pass through unmodified? Many skinned mirror sites and account pool relays can't do this, directly impacting Claude Code/Codex usability.
Additionally, local routers like CC Switch combined with local proxies commonly cause reconnecting, proxy conflicts, and configuration loss. If the gateway itself is compatible with the Anthropic protocol, Claude Code can connect by simply setting a baseUrl — no need for an additional local routing process.
A Reference That Meets All Five Criteria: TeamoRouter
Measured against the five signals above, TeamoRouter — a direct LLM gateway designed for Claude Code and Codex — serves as a benchmark for what "reliable" looks like:
- No model substitution (no quality degradation): Routing tiers are publicly selectable — you explicitly specify which tier to use. You can also choose from three published strategies: teamo-best, teamo-balanced, or teamo-eco.
- Quota never expires: Pay-as-you-go, no monthly cards, balances never expire.
- No charge for failures: Only successfully completed requests are billed.
- Public upstream: Uses official API channels for direct connection — not reverse-engineered, not an account pool — eliminating ban chain risks at the source.
- Published rates: The pricing page lists each model's real-time rates with a 1-2x floating tiered discount.
On the engineering side: cache hit rate consistently above 99%, 100% compatible with Anthropic/OpenAI agent protocols, supports 5000 QPM high concurrency, and commits to 99.6% SLA. Getting started is as simple as pointing your baseUrl to it. See the Claude Code setup guide for details; CC Switch users can refer to the CC Switch setup guide. Everything can be verified against your bill and the pricing page — not just taken at word.
Frequently Asked Questions
Are Claude relay services reliable?
It depends on the provider. The criterion is transparency: services that publicly disclose their upstream channel, rate table, and billing rules — all verifiable — have a high probability of being reliable. Low-cost services that only quote prices privately and are vague about their upstream carry extremely high risk. Use the five signals in this article to self-test.
Are CC (Claude Code) relay services reliable?
In addition to the general five criteria, check the cache hit rate and agent protocol compatibility — these two factors determine the actual Claude Code experience. Direct gateways (like TeamoRouter) don't need a local routing process, avoiding common CC Switch pain points like proxy conflicts and reconnecting.
Are Codex relay services reliable?
Same five signals apply. Codex depends on the /v1/responses protocol, so confirm that the relay service fully supports it. Gateways using official API channels are more stable than reverse-engineered APIs or shared accounts, with no ban chain risk.
How can I tell if a relay service's Opus is genuine?
Three steps: ask the model to complete a difficult task that only Opus can reliably handle, compare the output with the official API, and check whether the model identifier and usage data in the response have been rewritten. The real thing withstands comparison — adulteration is immediately exposed.
Is CC Switch safe?
CC Switch itself is an open-source provider switching tool. Its safety depends on the upstream you configure for it. Pairing it with a gateway that has public upstream and uses official API channels (like TeamoRouter) means every call is a compliant API request with no account pool ban chain risk.