[PERFORMANCE & BILLING] Allegro ($99) performs worse than free tier of competitors and March Moderato — suspected artificial throttling

What version of Kimi Code is running?

Kimi Code K2.7

Which open platform/subscription were you using?

Kimi Code (VS Code)

Which model were you using?

Kimi Code K2.7

What platform is your computer?

Windows 10 x64

What issue are you seeing?

Environment

• Extension: Kimi Code (VS Code)
• Version: K2.7
• Account: Paying subscriber
• Previous plan: Allegretto ($39/month) — purchased June 27, 2026
• Current plan: Allegro ($99/month) — upgraded July 3, 2026

Summary

The Allegro ($99) plan performs significantly worse than the Moderato plan from March 2026. Speed, quality, and credit efficiency have degraded with every update since February. Additionally, unused funds from the Allegretto plan were not refunded despite an email confirmation promising a refund.

Detailed Issues

1. Speed Degradation Over Time

Period	Plan	Speed	Quality
Feb–Mar 2026	Moderato (lower tier)	Fast	Good
Apr–May 2026	Moderato/Allegretto	Noticeably slower	Declining
Jun 2026	Allegretto ($39)	Slow	Poor
Jul 2026	Allegro ($99)	Slower than March Moderato	Worse than free tier of competitors

Observation: Performance degraded with each monthly update, contradicting advertised "x5–x15 credit improvements" in K2.7.

2. Credit Limits Do Not Match Advertised Claims

Claimed in K2.7 release notes: Credits increased x5 to x15.

Actual experience:

• Allegretto ($39) credits exhausted rapidly despite single-user usage
• Allegro ($99) does not provide proportional performance improvement
• A coding task that takes Kimi Code 1 hour on Allegro was completed by a free-tier competitor in 5 minutes with better quality

Conclusion: Advertised credit multipliers do not translate to real-world performance.

What steps can reproduce the bug?

3. Artificial Time Inflation / Fake Busywork

Reproducible behavior:

1. Agent completes a task (e.g., plan writing)
2. User observes the output file is ready and timestamps it
3. Agent continues generating reasoning, code, and "activity" for extended periods
4. Real completion time is inflated 2x–3x compared to actual output availability

Verification method:

• Close VS Code after output appears ready
• Reopen 10 seconds later
• Work is confirmed complete — proving the extended "reasoning" was unnecessary delay

Impact: Wastes credits on artificial token generation. The user pays for time the agent spends performing fake busywork.

4. Quality Regression

Examples:

• Code requiring 1 hour on Kimi Code Allegro was generated by a competitor's free tier in 5 minutes
• Even simple copy-paste tasks involve excessive error cycles
• Agent produces non-working solutions, then iterates with more non-working solutions
• Reasoning blocks are bloated with irrelevant content before any actionable output

5. Refund Not Processed

• June 27, 2026: Purchased Allegretto ($39)
• July 3, 2026: Upgraded to Allegro ($99) — unused Allegretto period remains
• Email confirmation: Received promise of refund for unused Allegretto funds
• As of [текущая дата]: Refund not received, no response to inquiry

Expected Behavior

• Allegro ($99) should perform at least as well as Moderato from March 2026
• Credits should last proportionally to plan tier
• Agent should stop processing immediately when output is ready (no fake busywork)
• Unused funds from tier switch should be refunded as confirmed via email

Actual Behavior

• Allegro ($99) is slower and lower quality than March Moderato
• Credits deplete faster than advertised
• Agent artificially extends session duration 2x–3x after output is ready
• Refund not processed despite email confirmation

What is the expected behavior?

Requests

1. Explain the performance regression from March Moderato to July Allegro with technical specifics
2. Verify whether artificial token generation / session inflation is occurring in K2.7
3. Process the refund for unused Allegretto funds immediately
4. Provide actual benchmark data comparing K2.7 speed/quality to pre-K2.7 versions
5. Publish transparent credit consumption metrics so users can verify "x5–x15" claims
6. Timeline for fix or compensation if performance cannot be restored

Additional information

No response

Start by reproducing the reported Kimi Code K2.7 behavior in VS Code using the described plan, timing, credit-consumption, and close/reopen checks. Compare the observed results with the March Moderato claims and the advertised K2.7 multipliers; the issue is done only when the regression, session behavior, and refund status are independently verified or disproved.