Benchmark · run 2026-06
Lapas vs Deepnest, measured on 37 nesting cases
Same parts, same plates, same time budgets: run head to head, then scored by an independent program that recomputes every number straight from the output geometry. No tool is trusted to grade its own work.
higher material density on average across the ten academic instances (86.6% vs 73.3%).
Lapas reaches a tighter nest in 10 seconds than Deepnest manages in five minutes, on all 4 shapes.
real sheet-metal jobs where Lapas fit the same parts on a whole sheet fewer. On the rest it tied, never worse.
Scorecard
Counting a case as a win only when the headline number (sheets used, then nest density) actually differs. Dead heats are shown as ties.
| Suite | What it measures | Lapas / tie / Deepnest |
|---|---|---|
| Density | best nest in fixed area | 10 / 0 / 0 |
| Real jobs | sheets used | 3 / 7 / 0 |
| Speed | density vs time budget | 12 / 0 / 0 |
| Scale | parts 50 to 2000 | 1 / 4 / 0 |
The benchmark's formal scorer breaks exact ties on the larger reusable remnant (a longer offcut is more useful), and by that stricter rule Lapas leads 36-0 (1 tie). We count those dead heats as ties here.
Density
ESICUP instances · 10Ten academic ESICUP strip-packing instances, the standard yardstick in nesting research. Both engines pack the same parts; higher density means a tighter nest.
Speed
density vs budgetHow tight a nest each engine reaches as the time budget grows from 10 seconds to 5 minutes. A higher, earlier-rising curve means a better nest, sooner.
Read the curves: Lapas (solid) sits above Deepnest (dashed) at every budget, and its 10-second point already clears Deepnest's 5-minute point.
Real jobs
3000 × 1500 mm · 10Ten synthetic sheet-metal jobs on a 3000 x 1500 mm plate: brackets, gussets, flanges, channels, tank panels. The metric a shop pays for is sheets consumed.
Across the whole batch Lapas used 25 sheets to Deepnest's 28, 3 fewer, while matching or beating density on every job.
| Job | Lapas (ours) | Deepnest | Result |
|---|---|---|---|
| brackets-mixed | 3 sh · 53.7% | 3 sh · 53.7% | tie |
| channels | 3 sh · 53.5% | 3 sh · 53.5% | tie |
| chevrons | 2 sh · 80.8% | 3 sh · 53.9% | −1 sheet |
| flanges | 3 sh · 53.8% | 3 sh · 53.8% | tie |
| gussets-dense | 2 sh · 80.9% | 3 sh · 53.9% | −1 sheet |
| long-rails | 3 sh · 53.5% | 3 sh · 53.5% | tie |
| mixed-batch | 2 sh · 81.1% | 3 sh · 54.1% | −1 sheet |
| plates-large | 2 sh · 78.3% | 2 sh · 78.3% | tie |
| small-parts | 3 sh · 53.8% | 3 sh · 53.8% | tie |
| tanks | 2 sh · 81.2% | 2 sh · 81.2% | tie |
Scale
50 → 2000 partsA fixed part mix scaled up to test how each engine holds up as part counts climb into the thousands. Fewer sheets is better.
How we tested
Methodology
Every case feeds identical inputs, the same parts, the same plate size, the same time budget, to both engines. Deepnest runs as deepnest-next v1.5.6 headless; Lapas runs its Lapas Core v1.0.31 engine.
We never trust a tool's own reported numbers. A separate scorer reads each engine's output geometry, recomputes density, sheet count and remnant from scratch, and validates that no parts overlap and spacing rules hold. A run that fails validation does not count.
Density suites use the public ESICUP strip-packing instances used in nesting research. Real-job, speed and scale suites use synthetic sheet-metal jobs built for this benchmark so the data can be published openly. Academic instances run three seeds (we report the median); large scale cases run a single seed.
Renders on this page are drawn by the scorer with one neutral style for both engines, so the picture you see is the geometry that was measured, not a marketing mock-up.
- Cases
- 37
- Run
- 2026-06 · June 16, 2026
- Lapas engine
- Lapas Core v1.0.31
- Deepnest
- deepnest-next v1.5.6
Don't take our word for it.
Upload one of your own DXF files and see the layout yourself. It's free and takes about two minutes.
Nest your own file for freeNo credit card · No install · Cancel anytime