TMB Scoreboard
“The cheapest model is the one that doesn’t create rework.” — The Monocle Bear
The TMB Scoreboard is how we decide which models CoeOS routes to. Not a leaderboard scraped from a paper: real, demanding tasks graded on strict rubrics, 42 tests across 17 axes, judged by Claude Opus 4.8 at temperature 0, totals computed mechanically.
The scoreboard lives on themonoclebear.com
Section titled “The scoreboard lives on themonoclebear.com”There is one scoreboard, and it is published and kept current there — live, sortable, with every entry and the full methodology.
What we measure
Section titled “What we measure”| Category | Axes |
|---|---|
| Coding | Python, General coding, Debugging, React and front-end, Swift, Refactoring |
| Planning | Task decomposition, Specs and handoff, Planning judgment |
| Reasoning | Reasoning and logic, Calculation, GDPR and compliance, Multi-jurisdiction legal |
| Writing | Creative writing, Business writing |
| Agentic | Agent execution, Agent safety |
The result is not “model X is good” but “model X scores 98 on debugging and 62 on Swift”: granular enough to route each request to the model proven best at that skill. CoeOS routes on 18 axes; the 18th, voice, is not part of the public scoreboard.
Cloud and local
Section titled “Cloud and local”The run covers cloud and local entries: the same model often appears twice, served by a cloud provider and on an OdyssAI-X cluster, so the cost of serving it at home is measured, not guessed.
Read next
Section titled “Read next”- Open the TMB Scoreboard →
- What CoeOS is → — how these scores become a routing table.