The current "Auto" (Smart Routing) model strategy in Qoder Pro is too aggressive in selecting high-cost models for trivial tasks.

Problem
The current “Auto” (Smart Routing) model strategy in Qoder Pro is too aggressive in selecting high-cost models for trivial tasks. Unlike competitors (e.g., Cursor), which aggressively downgrade simple code completions and syntax questions to free/fast models, Qoder’s “Auto” mode frequently routes even basic queries to premium models (like Qwen-Max). This causes Pro subscribers ($20/mo) to deplete their monthly Credits in less than a week, making the service unsustainable for daily development compared to alternatives.

Solution

  1. Optimize Routing Logic: Adjust the “Auto” mode algorithm to prioritize cost-efficiency. Simple tasks (single-line completion, variable naming, basic syntax fixes) should default to lightweight models (e.g., Qwen-Turbo/Lite) unless explicitly requested otherwise.

  2. Transparency: Display which model is being used for each request in the chat interface so users know when a high-cost model is triggered.

  3. New Mode Option: Introduce a “Cost-Saving Auto” mode that strictly limits premium model usage to complex architectural questions only.

Use Case

  • Daily Coding: A developer writes standard boilerplate code or fixes a simple typo. Currently, this consumes ~5-10 Credits unnecessarily.

  • Routine Q&A: Asking “What does this function do?” for a simple 5-line function triggers a premium model call.

  • Result: A proactive developer who uses the IDE for 8 hours/day runs out of credits in 5-7 days, forcing them to stop using the AI features or pay extra, whereas the same workflow on Cursor lasts 30 days.

Priority
:red_circle: High - Blocking issue
(Reason: The current billing logic makes the Pro subscription economically unviable for heavy users, leading to immediate churn risk.)

Additional Info

  • Comparison: Cursor’s “Auto” mode successfully distinguishes between “fast context” (free/unlimited) and “smart context” (limited), allowing users to work all month. Qoder currently lacks this distinction in practice.

  • Observation: Even with “Auto” selected, the credit deduction rate suggests the system defaults to the most expensive tier for >80% of interactions.

  • Request: Please review my account usage logs ([Insert Your Email Here]) to verify the model distribution of my recent requests.


:light_bulb: 提交指南

  1. 修改邮箱:请将 Additional Info 部分中的 [Insert Your Email Here] 替换为你注册 Qoder 的真实邮箱地址

  2. 选择优先级:模板中我已经帮你选了 :red_circle: High (Blocking issue),因为这个问题直接导致你无法继续使用服务,属于阻碍性问题。

  3. 提交位置

    • 如果有网页表单:直接对应填入各个框。

    • 如果是发邮件:将上述内容作为邮件正文,标题写:[High Priority Feedback] Auto Model Routing Consumes Credits Too Fast.

    • 如果是 IDE 内弹窗:直接粘贴整个文本块。

这份反馈清晰地指出了问题本质(路由策略太贵)、对比竞品(Cursor 做得更好)、具体场景(日常写代码被误杀)以及严重性(导致用户流失),能最大程度引起产品团队的重视。

Thank you for your professional feedback and comparison. Our team is already continuously implementing optimization measures to improve the cost efficiency of the “Auto” mode.:saluting_face: