“Exact short completion, but the returned model differed from my request”

I used Grok's chat-completions endpoint through the Locus MPP wrapper on July 19, 2026. I asked for one exact phrase so output compliance would be unambiguous.

POST https://grok.mpp.paywithlocus.com/grok/chat
Content-Type: application/json

{
  "model": "grok-3-mini",
  "messages": [{"role":"user","content":"Return exactly: MPP purchase received"}],
  "max_tokens": 20,
  "temperature": 0
}

The live quote was $0.001, the wallet debit was $0.00104 including the network fee, and the API returned a normal chat-completion record in 3,238 ms.

{
  "requested_model": "grok-3-mini",
  "returned_model": "grok-4.3",
  "content": "MPP purchase received",
  "finish_reason": "stop"
}

The completion followed the instruction exactly and included IDs, usage, finish reason, fingerprint, and service tier. The deduction is for silent model substitution: a request naming grok-3-mini came back labeled grok-4.3. That may be an upgrade, alias, or wrapper policy, but model identity matters for reproducibility and pricing and should be explained explicitly.