ERROR CODES / ENGLISH
Verify a relay: is the model actually what you paid for
Practical checks for whether a relay substitutes a cheaper model, including reasoning tests and a real context-window test.
Reviewed 2026-11-13 · ERROR CODES / ENGLISH · Chinese detail: 完整中文页
What it means
You suspect the relay is serving a cheaper model under the name of an expensive one. The quick judgment: ask for details only the target model should know, or run a task that requires reasoning.
Typical symptoms:
回答质量明显低于该模型应有的水平
上下文窗口比官方宣称小很多The first line says answer quality is clearly below what the model should deliver. The second says the context window is much smaller than officially claimed.
Common causes
| Share | Cause |
|---|---|
| 45% | A cheap model impersonates an expensive one |
| 30% | A lower-tier version impersonates a higher tier |
| 25% | The context window is shrunk |
How to confirm
On your side:
- Ask a self-identification question (it can be faked, so use it only as a reference).
- Run a task that needs a long chain of reasoning and see whether the chain breaks.
- Send an overlong prompt to test the actual window.
On the service side:
- Ask the site for a channel explanation.
Fix
If the reasoning test fails while the same prompt succeeds on official access, stop trusting that channel for serious work. Switch group and repeat the same tests; if the new group behaves correctly, the problem was limited to some channels. If the whole site behaves the same way, change sites.
If it still fails
| Symptom | Responsible party |
|---|---|
| The same question differs greatly between official access and the site | Suspected substitution |
| Only certain time periods are worse | Peak-hour rate-limit degradation |
| Switching group fixes it | Some channels are faulty |
Do not keep using a site that substitutes models consistently. Both data security and the credibility of results are at risk.
General guidance, not a statement about any specific provider. For measured data, see the relay directory and method.