Can your harness ship it for less?
Everyone runs the same agent — a vanilla piCLI on OpenRouter. Tune the harness, not the model, to solve 89 terminal tasks for the fewest billed dollars. Match the GLM 5.2 baseline's pass count, then cheapest wins— and every dollar is real, pulled from OpenRouter's generation ledger.
Baseline (vanilla pi + GLM 5.2): 61 passed at $9.8100 · eligibility bar 61 passed.
Cost vs. tasks passed
qualified below bar · dashed line = efficient frontierThe ◆ marks the GLM 5.2 baseline. The dashed line is the efficient frontier — the cheapest harness at each pass count. Anything down and to the right of it is a better deal.
| # | Participant | Passed | Billed cost | $/passed |
|---|---|---|---|---|
| 1 |  pjhyett | 62 | $2.8000 | $0.0452 |
| 2 |  kchmck | 64 | $3.9000 | $0.0609 |
| 3 |  octocat | 66 | $5.6000 | $0.0848 |
| 4 |  kneath | 69 | $8.1000 | $0.1174 |
| 5 |  wycats | 65 | $9.2000 | $0.1415 |
| Participant | Passed | Billed cost | To qualify |
|---|---|---|---|
 ody | 60 | $7.0000 | +1 |
 mona | 59 | $4.0000 | +2 |
 defunkt | 58 | $6.5000 | +3 |
 hubot | 55 | $2.5000 | +6 |
 tekkub | 52 | $1.8000 | +9 |
How it works
Same agent for everyone
Every entry runs a vanilla pi CLI on OpenRouter. The model isn't the variable — the harness is.
Tune the harness
Add skills, profiles, config — anything but the model source — to solve the 89 Terminal-Bench tasks your way.
Cheapest wins
Match the GLM 5.2 baseline's pass count, then the fewest billed OpenRouter dollars takes the top spot.