CLI returns 403 usage limit despite dashboard showing 0% — failed Login Device calls consuming 5h/weekly quota

What version of Kimi Code is running?

Kimi Code v0.36.1

Which open platform/subscription were you using?

managed:kimi-code

Which model were you using?

K3 thinking: max

What platform is your computer?

Ubuntu 24.04 LTS Server & Windows 11

What issue are you seeing?

Error: [provider.auth_error] 403 You've reached your usage limit for this billing cycle. Your quota will be
refreshed in the next cycle. To continue now, purchase extra usage or upgrade your plan:
https://www.kimi.com/code/#pricing

BUG: Rate-limit quota consumed by failed auth attempts; dashboard shows stale 0% while CLI shows 100%
Account: Allegretto plan, valid until 2026-08-21.
Summary: All Kimi Code CLI API calls return 403 You've reached your usage limit despite the web dashboard (kimi.com/code → My Quota) showing 51.52% total usage, 0% 5-hour usage, 0% 7-day usage, and 0% rate limit.
Contradiction:
Web dashboard / Code Console: 5-hour usage = 0%, weekly = 0%, rate limit = 0%
Windows CLI (fresh install, first-ever run): 5h limit = 100% used, weekly = 81% used
Ubuntu CLI (original machine): same 403 error
Trigger: Started after repeated kimi login / kimi continue / auth refresh attempts on Ubuntu. Each attempt failed with 403 usage limit or Invalid Authentication.
Usage History evidence: Code Console → Usage History shows 100 records (10 pages) of Login Device requests from kimi-code-cli/0.36.1, all returning 403. Sample request IDs:
65381863-eb2f-4786-bfe3-b1c14985207f (2026-08-18 01:00:17)
29303997-d255-47be-9b6c-271d64718a3f (2026-08-18 00:57:44)
cd2a6831-61b4-4c26-a89a-cd5f338f1bcf (2026-08-18 00:43:24)
Reproduction: Fresh Windows 10 install, never logged into Kimi Code before. Installed CLI, ran kimi login (succeeded), then kimi --yolo → immediate 403. CLI usage panel shows 5h limit already 100% consumed.
Hypothesis: Failed auth / login-device API calls are incorrectly counting against the 5-hour and weekly rate-limit quotas even though they return 403 and no actual LLM inference occurs. The web dashboard/Code Console cache does not reflect this consumption, creating a false "0% used" display while the real-time rate limiter blocks all requests.
Impact: Complete inability to use Kimi Code CLI on any device.

What steps can reproduce the bug?

Trying to use the CLI

What is the expected behavior?

The expected behavior is for it to work. I had not been using it to fill it up - yet I was being blocked.

Additional information

Start with the kimi login, kimi continue, auth refresh, and kimi --yolo flows, then trace how Login Device responses and quota errors are handled. Reproduce on a fresh installation and compare the CLI usage panel with the reported dashboard and Usage History; done means failed authentication attempts no longer consume quota or falsely block valid CLI use.