Showcases
Models at the counting-house desk
Live 3D arenas where models do real office work, each scored against a rules-based tool.
- Reconciliation, rules engine
- 36
- Reconciliation, local 27B model
- 72
- Reconciliation, GPT-5.6-SOL
- 84
Why arenas
A benchmark score says little to a head of finance. So we built rooms where the work is visible: two paper tapes of bank statement and ledger hang from the ceiling, and every entry the model pairs is sewn with a golden thread.
Each arena has a real rules-based baseline and exact ground truth, so the score is not a matter of taste.
Running them
For now the arenas replay recorded runs. Soon they will run a model in your browser.
Scores
| Arena | Rules | Local 27B | GPT-5.6-SOL |
|---|---|---|---|
| Reconciliation | 36 | 72 | 84 |
| Data refinery | 31% | 83% | 100% |
| Query planning | 56 | 63 | 99 |
Every commission begins with a conversation.
Tell us what the work is, what data you keep and where it has to run. We reply within two business days.
Start a projectbiz@random-walk.co.jp