Median lines of code across 5 tasks × 3 models × 5 runs. Lower is leaner.
Reproduce yourself: node benchmarks/run.js
· github.com/siddhamapple/needless
average(nums)Lines of code — median of 5 runs
All numbers
| Task | Haiku 3.5 | Sonnet 4.5 | GPT-4o mini | |||
|---|---|---|---|---|---|---|
| base | needless | base | needless | base | needless | |
| isPrime(n) | 19 | 3 -84% | 27 | 3 -89% | 22 | 4 -82% |
| formatPhone(s) | 31 | 4 -87% | 46 | 3 -93% | 38 | 4 -89% |
| truncate(s,n) | 11 | 2 -82% | 18 | 1 -94% | 13 | 2 -85% |
| shuffle(arr) | 22 | 6 -73% | 31 | 4 -87% | 26 | 6 -77% |
| average(nums) | 14 | 2 -86% | 21 | 1 -95% | 17 | 2 -88% |