MoonshotAI / MoonshotAI/kimi-code

repeated/stuttering words in thinking and output causing excessive token usage

Open
#2,047 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug
Dominant language
TypeScript
Stars
7.5k
Forks
1.2k
Avg merge
11h 53m
Merged PRs (30d)
350

Description

What version of Kimi Code is running?

0.28.1

Which open platform/subscription were you using?

oauth

Which model were you using?

kmi-k3

What platform is your computer?

Windows 11 23H2, build 10.0.22631.6199, x64

What issue are you seeing?

I am reporting a billing-impacting defect in Kimi Code. I downloaded Kimi Code only about 3 days ago and have been using it for about 1 day. I have not upgraded because this is a very recent install. Since the first day of use, Kimi Code has been generating the same word/phrase multiple times while thinking and also in the visible output text.
From my screenshot, examples include repeated fragments such as:
DesignDesignDesign
confirmed confirmed confirmed
Quick Quick Quick
filter filter filter
query query query
duplicated Grep / Shell tool blocks
repeated "Thinking" sections with duplicated text
This is not normal output quality; it looks like a streaming/rendering/model-output duplication bug. The repeated text has burned a large amount of my token usage, including thinking/output tokens. I am concerned that I am being charged for duplicated generated tokens caused by a product defect.

Image
What steps can reproduce the bug?
Steps to reproduce
  1. Open Kimi Code in a project folder.
  2. Run a normal coding prompt.
  3. Observe the "Thinking" output and tool/coding output.
  4. Repeated words/phrases appear, and tool blocks may duplicate.
Image
What is the expected behavior?
Expected behavior

Kimi Code should stream each token/block once. Thinking and final output should not contain repeated words like wordwordword unless explicitly requested.

Actual behavior

Kimi Code repeatedly emits duplicated words/phrases and duplicated tool/thinking blocks. This inflated output length and consumed excessive tokens.

Additional information

Please:
Confirm whether this is a known Kimi Code streaming/model-output duplication bug.
Confirm whether the repeated visible thinking/output tokens are billable.
Audit my recent usage for duplicated token generation.
Refund or credit the tokens consumed by this defect.
Advise a workaround and the fixed version/release when available.
Confirm that no further charges will accrue from this duplicated-output behavior.
I can provide a short screen recording or additional logs if needed.

adapter.log
main.log

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing adapter.log and main.log alongside the reproduction steps and screenshots. Trace whether repeated words and duplicated thinking or tool blocks originate in streaming, rendering, or model output, and determine whether the reported token usage is affected. Done means the duplication cause is identified and the expected single-token or single-block behavior is restored.

Written by the indexing model from the issue text.

Assessment

Tech stack
typescript
Domain
cli
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.