Your cheap LLM relay might be swapping the model. Here's how to catch it.

Cheap API relays for GPT, Claude, and DeepSeek may secretly swap flagship models for smaller, quantized, or context-truncated versions while continuing to bill for premium access. This degradation often occurs weeks after initial use, making it difficult to detect. To combat this, a verification playbook and an open-source tool have been developed to audit any OpenAI- or Anthropic-compatible endpoint. The method was shared by a professional working at daoxe to ensure gateway transparency.