← back to portfolio
Measurement · not a scoreboard
Token Burn
I run work across Claude Code, Codex, Hermes. Token usage isn't a leaderboard — it's a feedback loop for delegated intelligence: what work actually moved, whether the result got used, and whether the review burden was worth it. The hard part is honesty — you can't trust one token number across tools that count differently, so I keep three fidelity lanesseparate and never sum them into a single figure.
The three lanes (kept apart)
Exact tokens
329.7M
tokens · 542 receipts
Counted by the provider. Hard numbers.
Measured activity
2.7K
ops · 55 receipts
Real ops (tool calls, runs) — not tokens.
Estimate band
40K–90K
tokens · 1 receipts
Best-guess ranges. Never precise.
Three numbers, three confidence levels. Adding them would invent precision that doesn't exist.
Usage over time · daily exact-token total
2026-01-25peak 15M tok/day · exact lane · linear scale · hover for daily total2026-08-04
Context-mode savings · a different axis
384.9 MB
raw bytes kept out of the model's context window — context hygiene, not spend
96.2M
saved tokens (≈ token-equiv)
Where the work runs · by tool
Codex208
Hermes default183
Claude Code135
Gideon/Ollama71
ChatGPT1
bars = receipt count per tool (any lane)
Exact-token weight · by model
OpenAI Codex84.2M
claude-fable-581.3M
claude-opus-4-874.8M
claude-opus-524.8M
gpt-5.521.4M
gpt-5.6-sol15.1M
claude-haiku-4-5-202510018.1M
claude-opus-4-76.6M
claude-sonnet-55.3M
gpt-5.43.4M
anthropic2.4M
gideon:latest1.1M
gpt-5.3-codex594K
gpt-5.6-luna529K
gpt-5.2-codex20K
exact lane only · measured provider tokens
Per-agent burn · exact-token activity
Total
329.7M tok
Claude Code
200.9M tok
Hermes default
86.6M tok
Codex
41.1M tok
Gideon/Ollama
1.1M tok
lessmore · exact-token lane · shared log scale across agents
2026-01-25 → 2026-08-04 · Total plus one band per tool · log-scaled per day
Receipts · scrubbed
| Date | Tool | Model | Category | Lane | Value | Outcome | Next move |
|---|---|---|---|---|---|---|---|
| 2026-08-04 | Gideon/Ollama | gideon:latest | automation | exact | 5.2K | unrated | |
| 2026-08-04 | Gideon/Ollama | gideon:latest | automation | exact | 50K | unrated | |
| 2026-08-04 | Codex | gpt-5-codex | computer-work | activity | 5 | unrated | |
| 2026-08-04 | Claude Code | claude-opus-5 | agent-coding | exact | 2.0M | unrated | |
| 2026-08-04 | Claude Code | claude-opus-5 | agent-coding | exact | 256K | unrated | |
| 2026-08-04 | Codex | gpt-5.6-sol | computer-work | exact | 274K | unrated | |
| 2026-08-03 | Gideon/Ollama | gideon:latest | automation | exact | 15K | unrated | |
| 2026-08-03 | Gideon/Ollama | gideon:latest | automation | exact | 307K | unrated | |
| 2026-08-03 | Claude Code | claude-opus-5 | agent-coding | exact | 581K | unrated | |
| 2026-08-03 | Claude Code | claude-opus-5 | agent-coding | exact | 4.7M | unrated | |
| 2026-08-03 | Claude Code | claude-haiku-4-5-20251001 | agent-coding | exact | 540K | unrated | |
| 2026-08-03 | Hermes default | OpenAI Codex | agent-runtime | exact | 597K | used | monitor |
| 2026-08-03 | Hermes default | OpenAI Codex | agent-runtime | exact | 233K | used | monitor |
| 2026-08-02 | Gideon/Ollama | gideon:latest | automation | exact | 109 | unrated | |
| 2026-08-02 | Gideon/Ollama | gideon:latest | automation | exact | 400 | unrated | |
| 2026-08-02 | Gideon/Ollama | gideon:latest | automation | exact | 395K | unrated | |
| 2026-08-02 | Gideon/Ollama | gideon:latest | automation | exact | 724 | unrated | |
| 2026-08-02 | Gideon/Ollama | gideon:latest | automation | exact | 4.8K | unrated | |
| 2026-08-02 | Claude Code | claude-fable-5 | agent-coding | exact | 28K | unrated | |
| 2026-08-02 | Claude Code | claude-opus-5 | agent-coding | exact | 506K | unrated | |
| 2026-08-02 | Claude Code | claude-fable-5 | agent-coding | exact | 4.7M | unrated | |
| 2026-08-02 | Claude Code | claude-fable-5 | agent-coding | exact | 119K | unrated | |
| 2026-08-02 | Claude Code | claude-opus-5 | agent-coding | exact | 1.6M | unrated | |
| 2026-08-02 | Claude Code | claude-fable-5 | agent-coding | exact | 1.3M | unrated | |
| 2026-08-02 | Codex | gpt-5.6-sol | computer-work | exact | 130K | unrated | |
| 2026-08-02 | Hermes default | OpenAI Codex | agent-runtime | exact | 884K | used | monitor |
| 2026-08-02 | Hermes default | OpenAI Codex | agent-runtime | exact | 562K | used | monitor |
| 2026-08-02 | Hermes default | OpenAI Codex | agent-runtime | exact | 1.8M | used | monitor |
| 2026-08-02 | Hermes default | OpenAI Codex | agent-runtime | exact | 1.1M | used | monitor |
| 2026-08-02 | Hermes default | OpenAI Codex | agent-runtime | exact | 27K | used | monitor |
| 2026-08-01 | Gideon/Ollama | gideon:latest | automation | exact | 25K | unrated | |
| 2026-08-01 | Claude Code | claude-fable-5 | agent-coding | exact | 7.4M | unrated | |
| 2026-08-01 | Claude Code | claude-opus-5 | agent-coding | exact | 949K | unrated | |
| 2026-08-01 | Claude Code | claude-fable-5 | agent-coding | exact | 1.6M | unrated | |
| 2026-08-01 | Hermes default | OpenAI Codex | agent-runtime | exact | 2.3K | used | monitor |
| 2026-08-01 | Hermes default | OpenAI Codex | agent-runtime | exact | 596K | used | monitor |
| 2026-08-01 | Hermes default | OpenAI Codex | agent-runtime | exact | 33K | used | monitor |
| 2026-07-31 | Gideon/Ollama | gideon:latest | automation | exact | 8.3K | unrated | |
| 2026-07-31 | Codex | gpt-5-codex | computer-work | activity | 4 | unrated | |
| 2026-07-31 | Claude Code | claude-opus-5 | agent-coding | exact | 170K | unrated | |
| 2026-07-31 | Claude Code | claude-opus-5 | agent-coding | exact | 298K | unrated | |
| 2026-07-31 | Claude Code | claude-opus-5 | agent-coding | exact | 3.1M | unrated | |
| 2026-07-31 | Codex | gpt-5.6-sol | computer-work | exact | 52K | unrated | |
| 2026-07-31 | Codex | gpt-5.6-sol | computer-work | exact | 706K | unrated | |
| 2026-07-31 | Hermes default | OpenAI Codex | agent-runtime | exact | 563K | used | monitor |
| 2026-07-30 | Gideon/Ollama | gideon:latest | automation | exact | 8.9K | unrated | |
| 2026-07-30 | Claude Code | claude-fable-5 | agent-coding | exact | 3.4M | unrated | |
| 2026-07-30 | Claude Code | claude-fable-5 | agent-coding | exact | 96K | unrated | |
| 2026-07-30 | Claude Code | claude-haiku-4-5-20251001 | agent-coding | exact | 3.0M | unrated | |
| 2026-07-30 | Hermes default | OpenAI Codex | agent-runtime | exact | 1.2M | used | monitor |
| 2026-07-30 | Hermes default | OpenAI Codex | agent-runtime | exact | 589K | used | monitor |
| 2026-07-29 | Gideon/Ollama | gideon:latest | automation | exact | 9.2K | unrated | |
| 2026-07-29 | Claude Code | claude-fable-5 | agent-coding | exact | 3.7M | unrated | |
| 2026-07-29 | Claude Code | claude-fable-5 | agent-coding | exact | 73K | unrated | |
| 2026-07-29 | Claude Code | claude-fable-5 | agent-coding | exact | 611K | unrated | |
| 2026-07-29 | Claude Code | claude-sonnet-5 | agent-coding | exact | 4.4M | unrated | |
| 2026-07-29 | Hermes default | OpenAI Codex | agent-runtime | exact | 299K | used | monitor |
| 2026-07-29 | Hermes default | OpenAI Codex | agent-runtime | exact | 674K | used | monitor |
| 2026-07-28 | Gideon/Ollama | gideon:latest | automation | exact | 8.4K | unrated | |
| 2026-07-28 | Codex | gpt-5-codex | computer-work | activity | 3 | unrated |
Project, client and job detail are stripped at build time — only tool, model, generic category and outcome are shown. Snapshot generated 8/4/2026 · 764 source receipts.