Credit Consumption

Issue Description: Feedback report: 7afa87ea-b61b-425c-8b32-de7c2583c3f6 For last one week i am noticing that even project documentation is consuming more then 300 credits which was not the case earlier and this is really hurting my budget.

Steps to Reproduce: Ask the Qoder Desktop to update project documentation for example roadmap, changelog, memory, etc and you will notice that over 300 credits are gone. You can do this with any Project even just something which is purely documentation or technical.

Expected Behavior: Should consume less then 300 credits and it was only 50 - 100 credits before

Actual Behavior

Screenshots / Screen Recordings

Operating System

Current Qoder Version (Menu → About Qoder → Copy) Version 1.22.1

This is my fourth feedback report and i have already included the reference number also this time. I cannot keep on buying credits which is crazy and you guys need to either fix this or refund some of the credits. You can look at my consumption history. You will also see that for some basic tasks it has consumed over 2000 credits which is insane.

Hello,

After a deep analysis of this issue, here are our findings:

Where the credits actually went:
The high cost is not due to expensive single Markdown write operations. Instead, in Agent mode, a single instruction expands into multiple model calls. Taking your “Agree, proceed with commit” instruction at 10:30 as an example, the Agent actually modified 13 files (+291 lines / -57 lines), involving the Proposal, Architecture, Testing, Roadmap, Changelog, and CI workflow—while verifying and correcting itself step by step. This single instruction alone triggered 32 Ultimate model calls. The Ultimate model carries a 1.6x price multiplier, and each tool response requires carrying 100k+ tokens of context for the next request. While the cost per call is within the normal range, the total cost is high due to the compounding effect of “many calls × long context × high model multiplier.”

How to significantly reduce costs for similar tasks:

Use lower-multiplier models for documentation tasks.
The Performance model has a 1.1x multiplier, while the Efficient model is only 0.3x. Standard Markdown, changelog, or roadmap edits run well on the Efficient model, reducing costs by nearly an order of magnitude.

Start a new session for long-running tasks.
The history of the current session reached approximately 375 KB and triggered two context compactions. Starting a fresh session after completing a large task avoids carrying 100k+ tokens of history into every subsequent call.

Explicitly scope your requests.
For example: “Update only the CHANGELOG and ROADMAP; do not touch other files or add CI configurations.” This prevents the Agent from autonomously expanding the task to 13 files.

Lower the Reasoning Effort.
This session was set to “High” intensity. For routine documentation tasks, “Default” or “Medium” intensity is sufficient.

Recommendation:
For simple tasks, we suggest using our built-in low-multiplier models. If you have any further questions, please feel free to reach out.

Thanks for the detailed reply and this plus other issues with Qoder is making me move away from it.