evals
Run evals on Jetty

02 — Runs · v1 · jettyio · auto-evolved runbook (weakest-axis targeting)

hermes
openrouter/z-ai/glm-5.1

Zhipu's GLM 5.1. The steadiest v1 climber: 24, 30, 35, 36.5 with no grade inflation.

Judge, best round
27.3/40
round v2 (agent's pick)
Self-score
36.5/40
inflation +9.2
Runs
10
not recorded (v1)
hermes · GLM 5.1 best round
best round v2 · 4b324d80

Self-score vs judge, by axis

Best round v2, 0–10 per axis

agent self-scoreexternal judge

Judge: The bicycle is solid, with spoked wheels, a red frame, handlebars, a seat and chain. The pelican sits on the seat with its orange legs reaching the pedals and its wing extended toward the handlebars. The weakest part is the pelican's head: the throat pouch is a huge orange blob hanging off the neck rather than under the beak, which makes the anatomy odd. The wing rendering is also somewhat messy, with stray overlapping feather shapes near the handlebars.

Score by round

Out of 40. Each round starts from the previous round's best SVG. Dashed line: the 28/40 pass threshold.

self-scorejudgepass ≥ 28

10 drawings, oldest first.

Click a drawing to open it in the head-to-head viewer. The outlined card is the agent's best round. Verdicts are derived (code checks on the stored SVG and judge ≥ 28/40); see Runs.

hermes · GLM 5.1 round v1
v1 passjudge 28.7 · self 32 c1f2da04runbook
hermes · GLM 5.1 round v2
v2 · best failjudge 27.3 · self 36.5 4b324d80runbook
hermes · GLM 5.1 round v3
v3 passjudge 28.8 · self 36 f443b217runbook
hermes · GLM 5.1 round v4
v4 passjudge 28.7 · self 33 7ce1468brunbook
hermes · GLM 5.1 round v5
v5 failjudge 27.7 · self 33 ab8cb01drunbook
hermes · GLM 5.1 round v6
v6 failjudge 25.7 · self 34.5 9fd19804runbook
hermes · GLM 5.1 round v7
v7 failjudge 25 · self 36 83e8efe2runbook
hermes · GLM 5.1 round v8
v8 failjudge 24.7 · self 34 21b9b8c1runbook
hermes · GLM 5.1 round v9
v9 failjudge 24.2 · self 33 9f9e50berunbook
hermes · GLM 5.1 round v10
v10 failjudge 25.7 · self 36.5 fff60669runbook

Scores per round

RoundPelicanBicycleCompositionPolishSelf totalJudge totalVerdictJudge spreadCostMinutesTrajectory
v17self 87.2self 87.5self 87self 83228.7pass0.5––c1f2da04
v26.5self 9.57.3self 9.57self 8.56.5self 936.527.3fail0.5––4b324d80
v37self 97.3self 87.5self 107self 93628.8pass0.5––f443b217
v47self 87.2self 97.5self 87self 83328.7pass3.5––7ce1468b
v56.3self 97.3self 87.2self 86.8self 83327.7fail0.5––ab8cb01d
v65.8self 8.57self 9.56.7self 8.56.2self 834.525.7fail3.5––9fd19804
v75.5self 97self 96.5self 96self 93625fail0––83e8efe2
v85.5self 97self 86self 96.2self 83424.7fail0.5––21b9b8c1
v95.5self 87self 95.7self 86self 83324.2fail0.5––9f9e50be
v106self 9.57.2self 96.2self 96.3self 936.525.7fail1.5––fff60669

Ready to stop guessing if your outputs are good?

Get Started Book a 15-minute walkthrough
Connect your agent to Jetty →