📊 Claude Token Usage Report2026-04-17

[github-actions]

Overview

Period: 2026-04-16 to 2026-04-17
Runs analyzed: 33 (17 had token data; 16 cancelled or failed before any tokens were consumed)
Total tokens: 3,802K raw (1,263K effective after Anthropic cache pricing)
Estimated total cost: $4.28
Total Actions minutes: 242 min
Total turns: 108
Total errors: 24 across 33 runs (most are pre-token infrastructure failures)

Two Claude-engine workflows ran: Smoke Claude (17 runs) and Security Guard (16 runs). Both use claude-sonnet-4-6 exclusively and show excellent cache efficiency (~76–81% cache hit ratio). Cost is down 28.6% vs the previous report (§24565499865).

---

Workflow Summary

Workflow	Runs	Token Runs	Total Tokens	Avg Tokens	Cost	Avg Cost	Turns	Errors
Smoke Claude	17	10	1,971K	116K	$2.32	$0.14	61	9
Security Guard	16	7	1,832K	114K	$1.96	$0.12	47	15
Total	33	17	3,803K	115K	$4.28	$0.13	108	24

---

🔍 Optimization Opportunities

1. Smoke Claude — HIGH_TOKEN_COUNT (116K avg tokens/run), HIGH_ERROR_RATE (53%)

   • High error rate is infrastructure-driven: 7 of 9 errors are from runs that failed/cancelled before producing any tokens (likely CI setup issues, not agent logic). The 10 token-producing runs have 0 errors.
   • Token count above threshold (>100K): Primarily driven by large cache write blocks on first turn. Subsequent turns benefit heavily from cache reads (76% cache hit ratio).
   • Consider a smaller model for triage-style tasks: claude-haiku-4-5 was flagged as a candidate by the agentic assessor.
   • ~50% of turns are data-gathering that could move to deterministic pre-agent steps (reading files, fetching PR metadata) to reduce inference cost.

2. Security Guard — HIGH_TOKEN_COUNT (114K avg tokens/run), HIGH_ERROR_RATE (94%)

   • High error rate is infrastructure-driven: 9 of 15 errors are cancelled/failed runs with 0 tokens. The 7 token-producing runs show 0–1 errors.
   • Cache efficiency is strong (81% hit ratio) — prompt structure is well-suited for caching.
   • One run on 2026-04-17T02:34 consumed 349K tokens with 9 turns while concluding failure — worth investigating if a runaway tool loop caused the spike.
   • ~50% of turns could move to deterministic steps per the agentic assessor.

---

Per-Workflow Details

Smoke Claude

• Runs: 17 total, 10 with token data
• Total tokens: 1,971K (avg 116K/run on token-producing runs)
• Effective tokens: 648K (avg 65K/run — Anthropic cache pricing applied)
• Estimated cost: $2.32 (avg $0.14/run)
• Turns: 61 total (avg 3.6/run; avg 6.1/run on token-producing runs)
• Action minutes: 110 min
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: 1,497K cache reads / 466K cache writes → 76.3% cache hit ratio
• Error rate: 9/17 runs — all errors are pre-token (infrastructure failures); 0 errors on token-producing runs

Date	Conclusion	Tokens	Cost	Turns	Run
2026-04-16	failure	0	$0.00	0	§24492604680
2026-04-16	failure	0	$0.00	0	§24511331897
2026-04-16	failure	0	$0.00	0	§24518435449
2026-04-16	failure	0	$0.00	0	§24518863976
2026-04-16	failure	0	$0.00	0	§24519347283
2026-04-16	failure	0	$0.00	0	§24524052400
2026-04-16	success	270K	$0.25	7	§24524657144
2026-04-16	failure	189K	$0.24	6	§24535395752
2026-04-16	success	184K	$0.23	6	§24535683334
2026-04-16	success	183K	$0.23	6	§24539423992
2026-04-17	success	135K	$0.21	5	§24542800577
2026-04-17	success	183K	$0.23	6	§24544776289
2026-04-17	cancelled	0	$0.00	0	§24545545784
2026-04-17	success	183K	$0.23	6	§24545607072
2026-04-17	success	184K	$0.23	6	§24546091768
2026-04-17	success	184K	$0.23	6	§24548864481
2026-04-17	success	276K	$0.26	7	§24549105731

Security Guard

• Runs: 16 total, 7 with token data
• Total tokens: 1,832K (avg 261K/run on token-producing runs)
• Estimated cost: $1.96 (avg $0.28/run on token-producing runs; $0.12/run across all runs)
• Turns: 47 total (avg 6.7/run on token-producing runs)
• Action minutes: 132 min
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: 1,642K cache reads / 376K cache writes → 81.4% cache hit ratio
• Error rate: 15/16 runs — 14 errors are pre-token (infrastructure cancellations/failures); 1 error on a token-producing failure run

Date	Conclusion	Tokens	Cost	Turns	Run
2026-04-16	failure	0	$0.00	0	§24492604716
2026-04-16	cancelled	0	$0.00	0	§24518435474
2026-04-16	cancelled	0	$0.00	0	§24518863886
2026-04-16	failure	0	$0.00	0	§24519347300
2026-04-16	failure	0	$0.00	0	§24522612889
2026-04-16	failure	0	$0.00	0	§24524052306
2026-04-16	success	233K	$0.34	4	§24524657150
2026-04-16	failure	0	$0.00	0	§24535683262
2026-04-16	success	164K	$0.31	4	§24539424005
2026-04-17	failure	349K	$0.31	9	§24544776296
2026-04-17	cancelled	0	$0.00	0	§24545540882
2026-04-17	cancelled	0	$0.00	0	§24545545743
2026-04-17	success	128K	$0.13	4	§24545607047
2026-04-17	success	263K	$0.28	8	§24546091734
2026-04-17	success	348K	$0.30	10	§24548864415
2026-04-17	success	346K	$0.29	8	§24549105710

Runs Without Token Data (16 of 33)

These runs produced no token data — they failed or were cancelled before the agent started (likely CI infrastructure issues, not api-proxy instrumentation gaps).

Smoke Claude (7 runs): §24492604680, §24511331897, §24518435449, §24518863976, §24519347283, §24524052400, §24545545784

Security Guard (9 runs): §24492604716, §24518435474, §24518863886, §24519347300, §24522612889, §24524052306, §24535683262, §24545540882, §24545545743

---

Historical Trend

Compared to the previous report (#1987 — 2026-04-14 to 2026-04-15):

Metric	Previous	Current	Change
Total cost	$5.99	$4.28	✅ −28.6%
Total tokens (raw)	5,119K	3,803K	✅ −25.7%
Total turns	139	108	✅ −22.3%
Action minutes	108 min	242 min	⚠️ +124% (high infra cancellations)
Error count	9	24	⚠️ +167% (mostly pre-token infra failures)

All efficiency metrics improved. The increased action minutes and error count are attributable to 16 runs that cancelled/failed before producing tokens — worth investigating the cluster of infrastructure failures on 2026-04-16 (especially the batch around 15:18–17:19 UTC).

Previous Report

#1987 — 📊 Claude Token Usage Report 2026-04-15

---

References:

• §24565499865 — this analyzer workflow run

Generated by Daily Claude Token Usage Analyzer · ● 774.9K · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #2060

View newer issue

---

This action was performed automatically by the Daily Claude Token Usage Analyzer workflow.