LetTokenBurnDAYJOB: A Benchmark for Long-Horizon Professional Work
A benchmark of 130 long-horizon professional tasks in healthcare and finance requiring agentic workflows within containerized environments.
$ npm i -g lettokenburn-cli$ ltb bench info dayjob
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | 24.7 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| GPT-6 Astra | 11.6 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Claude Fable 5.1 | 9.6 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Claude Opus 5 | 8.4 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Grok-4.7 | 8.4 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Muse Spark 1.3 | 7.8 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
The DAYJOB benchmark consists of realistic requests paired with a workspace of files (PDFs, spreadsheets) and binary rubric criteria. Tasks are estimated to take professionals ~14-17 hours on average.
Scoring is based on an agentic judge verifying if all binary criteria in the task-specific rubric are met; attempts pass only if every criterion passes.