[BUG] Deceptive usage metrics for Claude Pro in VSCode

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

I asked Claude to write this summary so I could burn down my remaining 3% usage for the usage meter to reset. Not really why I am paying for this service!

Take it away, Claude:

Claude Code Pro Billing Feedback

Summary

The Claude Code Pro billing model presents usage as a time-based allowance (e.g. "5 hours per session") but the actual constraint is token volume. For the product's primary use case — software development — this creates a significant gap between what's advertised and what's delivered.

The core problem: time framing vs. token reality

Claude Code markets usage in hours, implying wall-clock time. In practice, a "5 hour" session can be consumed in 15-20 minutes of active coding work. This is because:

• Every conversation turn resends the full (growing) context history
• Reading source files pulls hundreds of lines into context per tool call
• Multi-file edits require multiple tool calls, each adding input and output tokens
• Context compaction (summarization) indicates the context window has already filled and reset, further accelerating token burn
• Opus, the flagship model, costs significantly more tokens per interaction than Sonnet

A typical real-world coding session — reading files, making edits across several modules, building, iterating — burns through the token budget at roughly 15-20x the rate the "hours" framing suggests.

The cooldown penalty for stopping

The session cooldown timer does not begin until the token budget is fully depleted. This means:

• If a user has 20% of a session remaining — not enough for meaningful work — they must either waste the remaining tokens or sit idle with no benefit
• Pausing to think, context-switch, or come back later is actively penalized
• The optimal strategy is to waste remaining tokens to trigger the reset timer sooner
• This is a perverse incentive that punishes thoughtful, efficient use of the tool

Effective weekly capacity

In practice, intensive coding sessions yield roughly 1-2 hours of actual wall-clock usage per week on the Pro plan, not the implied 35+ hours (5 hours/day x 7 days) that the time-based framing suggests. The combination of rapid token depletion and cooldown-gating makes the effective capacity a small fraction of what's advertised.

Recommendations

1. Present usage honestly — show token consumption or a normalized "compute units" metric rather than hours. Users can plan around an accurate budget; they can't plan around a misleading one.
2. Start cooldown on last activity, not on depletion — if a user stops interacting, begin the reset timer. Don't force them to drain remaining tokens.
3. Provide usage visibility — show real-time token consumption per conversation so users can make informed decisions about when to start fresh sessions, switch models, or reduce context.
4. Differentiate model costs transparently — if Opus costs 5x more than Sonnet per interaction, surface that at point of use, not buried in docs.

Context

This feedback comes from using Claude Code (Opus 4.6) for active development on a mid-sized Electron/TypeScript project (~10 source files, largest ~1700 lines). The work involved reading files, making targeted edits across multiple modules, building, and iterating — standard software engineering workflow, and exactly what Claude Code is designed for.

What Should Happen?

You already have a context usage indicator, you should have a session usage indicator. It is misleading to the point of feeling like bait-and-switch by stating a session is 5 hours when using the product for coding, which I believe is one of its key selling points, burns through that session in 15 minutes.

Error Messages/Logs

Steps to Reproduce

1. Start a new session in VSCode working on a coding project. Check the usage panel to see what remains in your session. Observe that it says "5 hours".
2. Work with Claude on coding tasks involving editing files and making code changes.
3. Observe that the usage panel shows the session is nearly entirely consumed after 15-20 minutes.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.92

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

This is using the VSCode plugin from Anthropic.