PulseAugur
EN
LIVE 20:41:36

Claude Code token counts questioned after deduplication reveals potential overstatement

A user attempting to reproduce a previous claim of 65.5 billion tokens used with Claude Code found a significant discrepancy in token counting. By analyzing a smaller subset of data, they discovered that deduplicating messages by ID resulted in approximately half the token count compared to summing all rows. This suggests that the original claim might have been inflated due to not properly handling duplicate message IDs, with cache reads accounting for a large portion of the token volume. AI

IMPACT Highlights potential issues in token usage reporting for LLMs, impacting cost calculations and performance assessments.

RANK_REASON User analysis of a specific product's usage data questioning previous claims.

Read on dev.to — LLM tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Claude Code token counts questioned after deduplication reveals potential overstatement

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Commentary
User analysis of a specific product's usage data questioning previous claims.
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
1 days old
Coverage has settled into its steady-state source set.

Full methodology in our editorial standards.

COVERAGE [1]

  1. dev.to — LLM tag TIER_1 Dansk(DA) · Michael Lip ·

    A Claude Code log sum doubled after message ID dedupe

    <p>I had a 65.5 billion-token Claude Code claim queued for publication. Then I tried to reproduce it.</p> <p>The old note called its result “30-day,” but its July 18 to August 19 dates cover 33 calendar days. The extractor and daily inclusion list are gone. In the files still on …