Daily spend bars, model side-by-side, top-5 expensive projects, and three paste-ready waste-cleanup findings — everything below is produced by the open-source codeburn CLI reading session files on the local disk. No API keys, no proxies, no remote service.
workspace-NNN;
all $ / token / call / session / one-shot-rate values are unmodified codeburn output (none hand-typed)
| Model | Cost | Calls | Edit turns | One-shot | $ / edit |
|---|---|---|---|---|---|
| Opus 4.7 | $3,340.91 | 25,068 | 361 | 71.2% | $9.25 |
| Opus 4.8 | $829.38 | 5,089 | 205 | 95.6% | $4.05 |
| Sonnet 4.6 | $194.05 | 3,983 | 28 | 82.1% | $6.93 |
| Haiku 4.5 | $45.45 | 3,587 | 98 | 51.0% | $0.46 |
| GPT-5.5 | $0.33 | 6 | 0 | — | $0.33 |
| Project | Date | Cost | Calls | Session |
|---|---|---|---|---|
| workspace-001 | 2026-06-03 | $106.46 | 445 | sess-3ff4d836 |
| workspace-002 | 2026-06-07 | $96.61 | 371 | sess-192682a9 |
| workspace-003 | 2026-06-03 | $95.24 | 465 | sess-f5d897e7 |
| workspace-004 | 2026-06-08 | $84.67 | 457 | sess-0e1a475f |
| workspace-006 | 2026-06-02 | $63.30 | 329 | sess-1d6df6a9 |
Claude made 3148 reads and 3242 edits (ratio 1.0:1). A healthy ratio is 4+ reads per edit. Editing without reading leads to retries and wasted tokens.
Potential savings · ~5.9M tokens (~$4.35)
Before editing any file, read it first. Before modifying a function, grep for all callers. Research before you edit.
Sessions with meaningful spend but weak delivery signals: workspace-001/sess-3ff4d836 on 2026-06-03: $106.46 (12 retries); workspace-002/sess-192682a9 on 2026-06-07: $96.61 (7 retries); workspace-003/sess-f5d897e7 on 2026-06-03: $95.24 (10 retries); workspace-004/sess-0e1a475f on 2026-06-08: $84.67 (20 retries, no one-shot edit turns); workspace-006/sess-1d6df6a9 on 2026-06-02: $63.30 (12 retries)
Potential savings · ~1823.8M tokens (~$1346.83)
Before continuing, name the deliverable in one sentence (PR title, file changed, command output you expect). Stop and check with me if (a) you spend more than 10 minutes without an edit, or (b) the same approach fails twice. Do not retry past two attempts on any single fix.
Effective input/cache tokens swamp output in these sessions: workspace-007/sess-3f4f7dbd on 2026-06-05: 9.7M effective input/cache vs 185.7K output (52.4:1); workspace-015/sess-034bc6c0 on 2026-06-08: 6.5M effective input/cache vs 145.0K output (44.9:1); workspace-005/sess-524c68a6 on 2026-06-01: 6.5M effective input/cache vs 172.5K output (37.5:1); workspace-031/sess-9fecf7f1 on 2026-06-05: 4.3M
Potential savings · ~79.6M tokens (~$58.75)
Start fresh before continuing. Use only the current goal, the relevant files, the failing command/output, and the constraints below. Restate the working context in under 10 bullets before editing.
Sessions costing more than 2x their peer-session average in the same project: workspace-051/sess-b003a67c on 2026-05-29: $13.43 (4.4x avg); workspace-062/sess-2b2698be on 2026-05-29: $9.12 (4.2x avg); workspace-064/sess-36599d9b on 2026-05-28: $8.26 (5.6x avg); workspace-033/sess-5c916e2a on 2026-05-24: $6.88 (2.1x avg); workspace-105/sess-456b17bf on 2026-05-28: $4.57 (8.6x avg); +33 more. These
Potential savings · ~69.7M tokens (~$51.48)
Before making changes, summarize the smallest viable plan. Keep context narrow, avoid broad searches, and stop after the first working patch so I can review before continuing.
762 redundant re-reads across sessions. Top repeats: courses.ts (355x), example-tasks.ts (82x), learn-courses.test.ts (54x). Each re-read loads the same content into context again.
Potential savings · ~457.2K tokens (~$0.338)