No model swapping: how to check which Claude model answered
“Model swapping” (sometimes called a silent downgrade) means you ask for one model but a cheaper, weaker one answers. 6090ti forwards each request to the model named in it, never swapped or downgraded. This article explains what that means in practice and how you can check it yourself.
How 6090ti handles it
- Forwarded to the model you choose. The model name in your request is the model the request goes to; it is not quietly replaced.
- An error, not a substitute. If the requested model is unavailable you get an error; another model is not used instead.
- Billed by the actual model. The model name, tokens and cost of every call are in the console usage records.
How to check
- Look at the model in the response. Anthropic-compatible responses include a
modelfield naming the model that answered:curl -s https://api.6090ti.com/v1/messages \ -H "x-api-key: <your key>" \ -H "anthropic-version: 2023-06-01" \ -H "content-type: application/json" \ -d '{"model":"<model name>","max_tokens":32,"messages":[{"role":"user","content":"hi"}]}' | grep -o '"model":"[^"]*"' - Compare with the usage records. Each entry in the console wallet’s request billing shows the model and cost, so you can match it against your requests.
- List the models your key can use. Request
https://api.6090ti.com/v1/models.
Why the same model sometimes answers differently
Model output varies from run to run and depends on the prompt, context length, max_tokens, temperature and other settings. Coding tools build very long contexts, and quality can drop when the context gets too long. If something looks clearly wrong, send support the time, model name and usage record of the request.
QQ user group: 1109931264
Related: Claude model APIs · model pricing