muse spark 1.3 contributor usage bug

Summary

I've used muse spark 1.2 contributor and it it'd been the cheapest model to use at command-code, today was charged my go plan and
started to solve some bugs on a project with 1.3 contributor, only with 54.1m tokens I hit my limits, it means I spent 3 credits on only 54m
tokens. The last month I did 1.2b tokens with 10 credits, doing some numbers it means the performance on my go plan is 666% worst. I hope I
get a reset or part of my credits back.

Expected Behavior

with 300k of cache input I was supposed to get 1 billion tokens on muse spark with 3 credits, I only got 55m.

Actual Behavior

Most of the conversation wasn't cached.

Steps to reproduce the issue

Just coding with muse spark 1.3 contributor

Command Code Version

1.46.0

Operating System

Windows

Terminal/IDE

Unknown

Shell

cmd.exe

Session file (optional)

6a9ce306-af51-4823-b044-d6e85a1eabc7.meta.json

command-code-session-6a9ce306.md

I couldn't upload the session as jsonl.

Fix prompt (optional)

No response

Additional context

No response

Start by reviewing the attached session metadata and session transcript, then reproduce the reported usage with Command Code 1.46.0, Muse Spark 1.3 Contributor, and the stated Windows cmd.exe environment. Compare cache accounting, token usage, and credit consumption with the expected behavior; done means identifying a reproducible product defect or documenting why the reported usage is expected.