Open WebUI triggers more Qwen tokens vs LM Studio

The "thinking" disparity between Open WebUI and LM Studio proves that the UI layer is just as critical as the model weights for reasoning performance.

• –Open WebUI's first-class support for `<think>` tags often includes hidden system instructions that explicitly demand detailed chain-of-thought processing.
• –When acting as a frontend for LM Studio or Ollama, Open WebUI may inject specific parameters like `thinking: true` that the native LM Studio UI suppresses or handles differently.
• –The observation highlights a "reasoning tax" where more verbose thinking leads to higher latency and token consumption without a guaranteed increase in output quality.
• –Developers using local interfaces must audit default system prompts to ensure they aren't unintentionally forcing models into expensive reasoning loops.