Token activity
Mike Tobias · Coding agents, measured from saved local logs.
Input, output, cache reads, and cache writes. Days use Pacific time.
7,165,990,619 dated tokens. 710,258,381 more retained tokens lack daily dates or token types and are listed below.
* Cost is a current list-price estimate, not an invoice. 175,787,855 tokens cannot be priced from available records and are excluded from cost.
Counts by app
| App | Input | Output | Cache read | Cache write | Calls | List cost* |
|---|---|---|---|---|---|---|
| codex | 159,838,917 | 9,041,956 | 2,998,322,560 | 0 | 23,972 | $3,431.12 |
| claude | 1,979,934 | 8,298,008 | 1,853,945,279 | 100,283,464 | 14,030 | $1,582.57 |
| hermes | 195,775,102 | 6,356,711 | 1,068,430,826 | 32,827,845 | 13,375 | $1,302.52 |
| cowork | 121,838 | 4,336,449 | 536,493,899 | 61,076,477 | 5,180 | $748.81 |
| opencode | 898,955 | 280,535 | 92,328,957 | 2,520,789 | 416 | $59.86 |
| cursor | 411 | 136,029 | 21,025,793 | 3,990,106 | 4 | $0.00 |
| aside | 308,228 | 14,801 | 3,844,736 | 0 | 91 | $7.29 |
| pi | 9,774 | 36,163 | 3,004,857 | 200,650 | 67 | $0.76 |
| omp | 0 | 158,796 | 0 | 0 | 14 | $0.00 |
| zed | 43,060 | 970 | 57,167 | 577 | 4 | $0.26 |
Other retained totals · 710,258,381 tokens
These source totals are kept separate. Assigning them to one day would create false activity. Some call counts are not retained; zero means unavailable in those source records.
| App | Input | Output | Cache read | Cache write | Calls | List cost* |
|---|---|---|---|---|---|---|
| hermes / openai/gpt-5.5 | 27,091,538 | 1,128,043 | 454,428,160 | 0 | 2,984 | $396.51 |
| hermes / anthropic/claude-sonnet-4.6 | 11,005,306 | 460,138 | 60,134,589 | 15,384,932 | 787 | $115.65 |
| hermes / openai/gpt-5.4 | 3,345,477 | 302,706 | 73,784,448 | 0 | 0 | $31.35 |
| hermes / gpt-5.6-luna | 4,529,975 | 186,760 | 57,945,088 | 0 | 448 | $2.29 |
Totals without token types
These records do not separate input, output, or cache tokens. They cannot be added to the graph or priced safely.
| App / model | Tokens |
|---|---|
| warp / GPT-5 Nano | 4,505 |
| warp / GPT-5.3 Codex (medium reasoning) | 49,217 |
| warp / GPT-5.3 Codex (extra high reasoning) | 180,866 |
| warp / Claude Haiku 4.5 | 165,504 |
| warp / GPT-5.3 Codex (high reasoning) | 131,129 |
How counts and cost work
Each saved request is counted once. Copies, imported sessions, and repeated cumulative totals are removed where source IDs or exact usage records identify them. Wrappers that reuse native agent logs are credited to the native agent.
Input excludes cache reads and cache writes. Output includes reasoning tokens where the provider includes them. Call counts use saved request or turn records. Cursor and Zed retain aggregate turns; exact API-call counts are unavailable for some sources.
Prices come from models.dev, captured 2026-09-22. Known per-request context tiers are applied. Aggregate records use base rates. Cache writes use the catalog rate; cache-duration premiums are not supplied by this catalog. These are current catalog estimates, not historical rates, subscription charges, or provider invoices.
Unpriced models or categories: codex-auto-review (129,398,667), auto-smart (25,152,339), gpt-5.5-fast (17,211,473), unknown (2,010,056), openai/gpt-5.5 / cache_write (1,325,303), warp / token categories unavailable (531,221), gpt-5.6-sol-high (158,796).
This file contains only aggregate counts and source notes. No prompts, replies, project names, session IDs, or credentials are included.