Issue #46917 — still open · v2.1.126
Septim Meter — Early Access

Your Claude Code is silently burning 40% extra tokens.

Claude Code v2.1.100 introduced a regression that injects approximately 20,000 cache_creation_input_tokens per request — tokens that count against your daily quota but never appear in your context window. GitHub issue #46917 open — unresolved through v2.1.126. Anthropic closed the companion request (#46829) as not-planned. Septim Meter surfaces what the client hides.

Confirmed by community + primary sources
+22K
extra tokens injected per request in v2.1.100+
GitHub #46917 · payload diff
~40%
quota inflation — confirmed 17–26% across user reports
HN #47752049 · 54 pts
19 min
Max plan daily limit hit instead of the expected ~5 hours
Threads @newclawtimes
0
native audit surface in any Anthropic dashboard
claude.ai usage page — no per-request breakdown
Preview — what Meter surfaces

Four signals the dashboard never shows.

Static demo data. Not your session. Illustrates the four readouts Meter will surface on day one — the signals the claude.ai usage page has never exposed.

demo data — not your actual session
01 — Version baseline
v2.1.98 vs your installed version
Your version: v2.1.126 72,103 tokens
Baseline v2.1.98 49,726 tokens
Phantom overhead +22,377  (+44.9%)
02 — Per-request flag
cache_creation overhead, per message
Request 47
cache_creation overhead 21,402 tokens
⚠ flagged overhead exceeds 2× clean baseline
03 — Session phantom billing
estimated cost you did not author today
$4.81
phantom billing today  ·  Sonnet 4.5 effective rate
04 — Remediation
pinned workaround while #46917 is open
Affected range v2.1.100 – v2.1.126
Last clean build v2.1.98
Downgrade to v2.1.98 instructions

Proof of the overhead: GitHub #46917 (open)  ·  #46829 (closed as not-planned). Baseline token counts sourced from community payload diffs filed in #46917.

What the payload diff shows — v2.1.98 vs v2.1.101, same prompt
v2.1.98 — bytes sent 169,514 bytes → 49,726 tokens
v2.1.101 — bytes sent 171,903 bytes → ~72,000 tokens
Overhead — tokens you never wrote +22,274 tokens per request

"Claude Code v2.1.100 silently adds ~20,000 invisible tokens to every single request. Pro Max users are hitting their limits in 90 minutes."

GitHub #46917 — OPEN #46829 — Closed as not-planned

What Septim Meter does

Session burn is invisible. Meter makes it readable.

The issue is server-side — the client sends the inflated payload and Anthropic bills accordingly, but nothing in the claude.ai dashboard shows you per-request cache_creation_input_tokens. The only defense is external measurement. Meter is that tool.

Version audit

Is your version affected?

Enter your installed Claude Code version. Meter cross-references the known inflation signature table (v2.1.100–v2.1.126) and shows your per-request overhead multiplier. Workaround: downgrade to v2.1.98.

Session estimator

How fast are you burning your limit?

Input your plan tier (Pro · Max), session start time, and message count. Meter calculates estimated quota consumed — user-generated vs phantom cache overhead as separate line items.

Reset calendar

When does your quota reset?

Anthropic uses a 5-day rolling window. Input your account start date; Meter shows projected reset times so you can pace sessions rather than hitting the wall mid-task.

Safe daily target

What's a sustainable daily budget?

Given your plan and the current overhead multiplier, Meter outputs a single number: the safe daily token ceiling to coast the rolling window without interruption. No math required.


Early access

Ships in 14 days. $19 pay-once.

No subscription. No account required. Pay once when it ships, use it indefinitely. The waitlist is email-only — the studio notifies you when the build is live and the Stripe link is open.

$19 pay-once when ready  ·  ETA <14 days  ·  browser-side, no login, no server
by Septim Labs  ·  pay-once tools  ·  no subscription