📊 Claude Token Usage Report2026-05-01

[github-actions]

Overview

Period: 2026-04-30T23:26Z to 2026-05-01T17:28Z
Runs analyzed: 42 (35 had token data; 7 had 0 tokens due to early failures)
Total tokens: 7,847K (effective: 2,419K with caching)
Estimated total cost: $6.14
Total Actions minutes: 201 min

⚠️ 7 runs had no token data (3× Security Guard, 4× Smoke Claude) — early failures before the agent started.

Workflow Summary

Workflow	Runs	Total Tokens	Avg Tokens/Run	Total Cost	Avg Cost/Run	Avg Turns
Security Guard	20	5,168K	258K	$4.98	$0.25	6.6
Smoke Claude	22	2,679K	122K	$1.16	$0.05	5.2

🔍 Optimization Opportunities

1. Security Guard — $0.25/run avg, high token count, elevated error rate ⚠️

   • High avg tokens (258K/run, threshold 100K): claude-sonnet-4-6 is handling extensive code analysis; consider whether all review passes are necessary or can be parallelized/cached in deterministic pre-steps
   • High error rate: 8/20 runs (40%) had errors — above the 30% threshold; investigate reliability issues
   • Rising cost trend: +32% avg cost/run vs previous report ($0.19 → $0.25) — exceeds 20% increase threshold
   • Some runs reached 17–18 turns (threshold: 15); the highest-cost run ($0.61, 17 turns) suggests the agent is doing significant back-and-forth
   • Cache efficiency is 86.1% (healthy) on claude-sonnet-4-6 — caching is working well; the cost is driven by volume

2. Smoke Claude — cost stable and low at $0.05/run ✅

   • Uses claude-haiku exclusively — appropriate model choice
   • Cache efficiency 75.2% — reasonable for short smoke test conversations
   • 5/22 runs had errors or missing data (23%, below threshold)
   • No action needed

Per-Workflow Details

Security Guard

• Runs: 20 (links below)
• Total tokens: 5,168K (avg 258K/run)
• Estimated cost: $4.98 (avg $0.25/run)
• Turns: 132 total (avg 6.6/run)
• Actions minutes: 103
• Model breakdown: claude-sonnet-4-6 (primary: 96 requests, 4.4M cache read, 706K cache write) + claude-haiku-4-5-20251001 (12 requests, orchestration)
• Cache analysis: cache_read=4,377K, cache_write=706K, efficiency=86.1% — writes are well-reused
• Error rate: 8/20 runs (40%) ⚠️

Date	Run ID	Conclusion	Tokens	Cost	Turns
2026-05-01	§25224916137	success	639K	$0.61	17
2026-05-01	§25224279505	success	589K	$0.57	16
2026-05-01	§25223939926	failure	526K	$0.48	11
2026-05-01	§25220381062	success	496K	$0.49	18
2026-05-01	§25200682002	success	415K	$0.39	9
2026-05-01	§25200727240	cancelled	455K	$0.26	11
2026-05-01	§25199258915	success	633K	$0.62	12
2026-05-01	§25201004443	success	386K	$0.38	8
2026-05-01	§25218663410	success	284K	$0.33	8
2026-05-01	§25200435610	success	283K	$0.26	8
2026-05-01	§25217346225	success	237K	$0.32	8
2026-05-01	§25200380157	success	225K	$0.27	6
2026-04-30	§25194366508	failure	0	$0.00	0
2026-05-01	§25195690726	success	0	$0.00	0
2026-05-01	§25204711247	failure	0	$0.00	0
2026-05-01	§25214713163	success	0	$0.00	0
2026-05-01	§25218522054	failure	0	$0.00	0
2026-05-01	§25218590968	success	0	$0.00	0
2026-05-01	§25219394037	failure	0	$0.00	0
2026-05-01	§25221711163	failure	0	$0.00	0

Smoke Claude

• Runs: 22
• Total tokens: 2,679K (avg 122K/run)
• Estimated cost: $1.16 (avg $0.05/run)
• Turns: 114 total (avg 5.2/run)
• Actions minutes: 98
• Model breakdown: claude-haiku-4-5-20251001 exclusively (80 requests)
• Cache analysis: cache_read=1,996K, cache_write=660K, efficiency=75.2%
• Error rate: 5/22 runs with errors or no data (23%)

Runs Without Token Data

These 7 runs had token_usage = 0 — agent likely didn't start due to infrastructure or pre-agent failures:

• Security Guard: §25194366508 (failure), §25204711247 (failure), §25218522054 (failure), §25219394037 (failure), §25221711163 (failure) — plus 2 success runs with 0 tokens
• Smoke Claude: §25221711162 (failure) + additional zero-token runs

Historical Trend

Compared to previous report (#2325 — 2026-04-30):

Workflow	Prev Avg Cost	Curr Avg Cost	Change
Security Guard	$0.19/run	$0.25/run	+32% ⚠️
Smoke Claude	$0.06/run	$0.05/run	-17% ✅

Security Guard crossed the +20% increase threshold. The increase is driven by more PRs with larger diffs being reviewed — the recent copilot/optimize-api-proxy-efficiency PR branch contributed several high-turn runs. Cache efficiency remains excellent, so the cost growth reflects genuine workload increase rather than caching regression.

Previous Report

#2325 — 2026-04-30

References:

• §25231072523 (this analysis run)

Generated by Daily Claude Token Usage Analyzer · ● 395.8K · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #2407

View newer issue

---

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