← Back to leaderboard  ·  Show all task instances

johnkerl/miller

Miller is like awk, sed, cut, join, and sort for name-indexed data such as CSV, TSV, and tabular JSON

9,842 go hard
14,637
Generated Behavioral Tests
38.9%
Best Score
GPT-5.6 Sol (xhigh)

Results by Model

Hover for details · Click a model to open it · Drag to zoom the x-axis · Shift-drag to pan · The line marks the Pareto frontier (best score per cost)

Per-Model Results

25 runs
# Model Score Cost Calls
1 GPT-5.6 Sol (xhigh) 38.9% $5.60 36 trace →
2 GLM-5.2 29.2% $37.93 347 trace →
3 Claude Opus 4.8 (xhigh) 28.8% $22.19 192 trace →
4 Gemini 3.7 Flash 28.2% $1.48 113 trace →
5 Muse Spark 1.1 (xhigh) 25.7% $1.06 202 trace →
6 GPT-5.6 Sol 25.1% $1.33 21 trace →
7 Claude Opus 4.7 (xhigh) 24.1% $8.12 127 trace →
8 GPT 5.5 (high) 23.4% $3.02 26 trace →
9 Claude Opus 4.6 22.9% $13.86 322 trace →
10 Muse Spark 1.2 (xhigh) 22.1% $0.63 131 trace →
11 Claude Opus 4.7 19.0% $2.86 62 trace →
12 Gemini 3.6 Flash 18.5% $3.97 157 trace →
13 GPT 5.5 16.9% $1.02 15 trace →
14 Gemini 3.5 Flash 11.5% $5.31 174 trace →
15 Gemini 3 Flash 11.5% $0.20 58 trace →
16 GPT 5.4 8.0% $0.32 11 trace →
17 Muse Spark 1.3 (xhigh) 7.5% $1.93 226 trace →
18 Muse Spark 1.3 (max) 4.9% $6.58 369 trace →
19 Gemini 3.1 Pro 4.9% $0.95 59 trace →
20 Claude Haiku 4.5 4.9% $1.61 206 trace →
21 GPT 5.4 mini 4.7% $0.02 8 trace →
22 GPT 5 mini 3.4% $0.02 12 trace →
23 Claude Opus 5 (xhigh) 0.0% $7.60 104 trace →
24 GPT 5.5 (xhigh) 0.0% $5.15 60 trace →
25 Claude Sonnet 4.6 0.0% $5.77 286 trace →

Click a row to replay how that model rebuilt this program, or the model name to open its full run