Benchmark · 16 October 2026

Router comparison: method and full results

Five open-source PCB autorouters, plus the hosted Zap!, on 26 routing jobs across 11 public KiCad boards. This page has everything behind the article: how runs were judged and scored, the machine and settings, and the result of every case.

5open-source routers

11public KiCad boards

26cases in 4 tasks

390scored runs

2audits

Watch every case

Pick a task and a board. Each router's copper appears over its real route time, beside its vias, track length and verdict. Click any case in the tables below to load it here.

Every case

A case is valid for a router when all three of its runs pass. Green cells are valid; red cells show what went wrong. Each cell gives the median route time, vias and track length over the three runs. Zap!'s results are provisional and use the repeats checked so far. Click a case to watch it.

Whole board

Route every connection from placed parts. All tracks, vias and copper zones removed first.

drandyhaas/KiCadRoutingTools 6/6 valid	TraceMaker 6/6 valid	fastroute 6/6 valid	fab_router 3/6 valid	Freerouting 5/6 valid	Zap! 6/6 valid
valid 3/3 1.9 s 5 vias · 377 mm	valid 3/3 1.3 s 0 vias · 354 mm	valid 3/3 0.7 s 0 vias · 367 mm	valid 3/3 0.7 s 12 vias · 353 mm	valid 3/3 3.1 s 0 vias · 367 mm	valid 2/2 4.3 s remote
valid 3/3 2.2 s 15 vias · 738 mm	valid 3/3 5.9 s 4 vias · 717 mm	valid 3/3 0.8 s 1 vias · 755 mm	2 open 0/3 9.8 s	valid 3/3 3.5 s 0 vias · 749 mm	valid 3/3 6.9 s remote
valid 3/3 4.8 s 51 vias · 1519 mm	valid 3/3 57 s 34 vias · 1458 mm	valid 3/3 1.2 s 35 vias · 1529 mm	3 open 0/3 1.9 s	valid 3/3 6.2 s 27 vias · 1575 mm	valid 2/2 8.9 s remote
valid 3/3 5.7 s 27 vias · 853 mm	valid 3/3 18 s 23 vias · 802 mm	valid 3/3 1.4 s 9 vias · 903 mm	valid 3/3 1.2 s 59 vias · 828 mm	valid 3/3 7.2 s 5 vias · 944 mm	valid 2/2 6.7 s remote
valid 3/3 4.3 s 30 vias · 764 mm	valid 3/3 33 s 33 vias · 824 mm	valid 3/3 1.6 s 18 vias · 859 mm	5 open 0/3 3.5 s	valid 3/3 9.7 s 16 vias · 847 mm	valid 2/2 7.7 s remote
valid 3/3 2.6 s 7 vias · 416 mm	valid 3/3 2.5 s 4 vias · 397 mm	valid 3/3 0.8 s 3 vias · 449 mm	valid 3/3 0.7 s 27 vias · 387 mm	1 open 0/3 4.7 s	valid 2/2 4.3 s remote

Reroute after an edit

On a routed board, remove the copper of one net (small), all nets of one part (medium) or a connector-sized group (large), then reconnect it without touching anything else.

drandyhaas/KiCadRoutingTools 7/9 valid	TraceMaker 5/9 valid	fastroute 3/9 valid	fab_router 6/9 valid	Freerouting 2/9 valid	Zap! 5/9 valid
valid 3/3 3.2 s 13 vias · 1546 mm	valid 3/3 5.2 s 3 vias · 1496 mm	1 open 0/3 1.7 s	valid 3/3 2.7 s 41 vias · 1522 mm	1 open 0/3 6.2 s	valid 2/2 6.9 s remote
valid 3/3 1.5 s 1 vias · 1505 mm	valid 3/3 1.5 s 0 vias · 1493 mm	valid 3/3 0.7 s 0 vias · 1494 mm	valid 3/3 0.9 s 1 vias · 1484 mm	valid 3/3 3.2 s 0 vias · 1494 mm	valid 2/2 4.9 s remote
valid 3/3 1.6 s 0 vias · 1502 mm	valid 3/3 1.3 s 0 vias · 1493 mm	1 open 0/3 0.9 s	valid 3/3 0.7 s 2 vias · 1498 mm	1 open 0/3 4.3 s	valid 2/2 5 s remote
1 error 0/3 9.8 s	1 error 0/3 9.1 s	1 open 0/3 1.7 s	1 open 0/3 0.9 s	2 open 0/3 8.9 s	1 open 0/2 22.8 s remote
valid 3/3 3.2 s 21 vias · 480 mm	2 open 0/3 1.6 s	valid 3/3 0.7 s 14 vias · 491 mm	valid 3/3 0.8 s 20 vias · 473 mm	valid 3/3 3.5 s 14 vias · 491 mm	2 open 0/2 15.6 s remote
1 error 0/3 2.9 s	1 error 0/3 1.3 s	1 error 0/3 0.7 s	2 errors 0/3 0.7 s	1 error 0/3 3.5 s	valid 2/2 13.7 s remote
valid 3/3 3.6 s 27 vias · 520 mm	valid 3/3 1.6 s 31 vias · 502 mm	2 open 0/3 1.2 s	valid 3/3 0.9 s 26 vias · 562 mm	1 open 0/3 4.8 s	valid 2/2 7.1 s remote
valid 3/3 2.1 s 23 vias · 533 mm	valid 3/3 1.3 s 23 vias · 533 mm	valid 3/3 0.7 s 23 vias · 534 mm	valid 3/3 0.7 s 27 vias · 532 mm	1 open 0/3 4.7 s	invalid 0/2 6.5 s remote
valid 3/3 2.0 s 23 vias · 533 mm	2 open 0/3 2.1 s	1 open 0/3 0.9 s	1 open 0/3 3.1 s	1 open 0/3 5.5 s	2 open 0/2 6.5 s remote

Power widths

Route the whole board while giving power nets their required track widths.

drandyhaas/KiCadRoutingTools 4/5 valid	TraceMaker 4/5 valid	fastroute 4/5 valid	fab_router 2/5 valid	Freerouting 4/5 valid	Zap! 5/5 valid
valid 3/3 2.1 s 5 vias · 378 mm	valid 3/3 1.3 s 0 vias · 354 mm	valid 3/3 0.7 s 0 vias · 369 mm	valid 3/3 0.7 s 12 vias · 354 mm	valid 3/3 3.1 s 0 vias · 367 mm	valid 2/2 4.4 s remote
valid 3/3 3.4 s 12 vias · 737 mm	valid 3/3 5.7 s 5 vias · 707 mm	valid 3/3 0.7 s 0 vias · 751 mm	2 open 0/3 12 s	valid 3/3 3.4 s 0 vias · 749 mm	valid 3/3 7.7 s remote
valid 3/3 5.0 s 8 vias · 788 mm	valid 3/3 6.2 s 0 vias · 794 mm	valid 3/3 1.0 s 0 vias · 826 mm	valid 3/3 1.9 s 40 vias · 845 mm	valid 3/3 3.5 s 0 vias · 829 mm	valid 2/2 6.7 s remote
valid 3/3 7.0 s 29 vias · 1634 mm	valid 3/3 12 s 2 vias · 1529 mm	valid 3/3 1.2 s 2 vias · 1557 mm	3 open 0/3 6.6 s	valid 3/3 5.5 s 5 vias · 1558 mm	valid 3/3 9.2 s remote
1 error 0/3 5.3 s	3 errors 0/3 2.6 s	1 open 0/3 1.1 s	1 open 0/3 1.6 s	6 open 0/3 5.5 s	valid 2/2 5.2 s remote

Keep a ground pour

Route the whole board without changing its existing ground pours; the pour must stay connected.

drandyhaas/KiCadRoutingTools 3/6 valid	TraceMaker 3/6 valid	fastroute 3/6 valid	fab_router 2/6 valid	Freerouting 1/6 valid	Zap! 3/6 valid
valid 3/3 2.4 s 4 vias · 252 mm	valid 3/3 1.5 s 0 vias · 244 mm	valid 3/3 0.8 s 0 vias · 248 mm	valid 3/3 0.8 s 9 vias · 240 mm	valid 3/3 3.2 s 0 vias · 252 mm	1 error 0/2 5 s remote
10 errors 0/3 9.6 s	valid 3/3 34 s 7 vias · 481 mm	valid 3/3 1.3 s 3 vias · 578 mm	6 open 0/3 14 s	2 open 0/3 6.5 s	valid 2/2 9.4 s remote
valid 3/3 5.4 s 28 vias · 497 mm	2 open 0/3 11 s	2 open 0/3 1.3 s	1 error 0/3 1.0 s	1 open 0/3 7.2 s	1 open 0/2 13.8 s remote
1 error 0/3 3.0 s	1 open 0/3 2.8 s	1 open 0/3 0.8 s	valid 3/3 0.8 s 14 vias · 224 mm	1 open 0/3 3.6 s	valid 2/2 5.8 s remote
3 open 0/3 45 s	6 open 0/3 48 s	9 open 0/3 7.1 s	15 open 0/3 44 s	10 open 0/3 15 s	5 open 0/2 64.5 s remote
valid 3/3 2.6 s 6 vias · 346 mm	valid 3/3 3.1 s 17 vias · 264 mm	valid 3/3 0.9 s 15 vias · 295 mm	1 open 0/3 1.2 s	1 open 0/3 4.3 s	valid 2/2 4.9 s remote

Ranking and scores

Ranked by valid cases, then Efficient, then Fast. Scores are out of 100, computed per case among routers with a valid board, then averaged over each router's valid cases where at least two routers were valid. Each router's value on a case is its median over three runs.

Router	Valid cases	Efficient	Fast	Fast (CPU time)	Busy cores
drandyhaas/KiCadRoutingTools	20/26	70	32	24	0.85
Zap!*	19/26	—	—	—	—
TraceMaker	18/26	87	30	15	3.07
fastroute	16/26	91	98	95	1.51
fab_router	13/26	66	93	93	0.94
Freerouting	12/26	95	21	4	3.54

Efficient

50 × (fewest vias + 1) / (vias + 1) + 50 × shortest length / length

The router with the fewest vias and shortest track on a case scores 100.

Fast

100 × fastest time / time

Full path from KiCad input to KiCad output, including any file conversion. Twice as slow as the fastest scores 50.

Fast (CPU time)

100 × least CPU seconds / CPU seconds

Same, counting CPU time sampled every 100 ms, so extra cores do not help. Not part of the ranking.

* Zap! is hosted and closed-source. Provisional: 55 of 78 runs checked, paused at the service's daily job limit. It has no scores: its time is measured on a remote server, and it was judged with checker v4.4, which also allows router-added pours on boards that start with none. Rechecking all 390 open-source runs with v4.4 changed no verdict. On the same 55 runs: drandyhaas/KiCadRoutingTools 43 valid, Zap! 41, TraceMaker 39, fastroute 35, Freerouting 27, fab_router 26.

How runs are judged

Connected

No missing connections, measured by KiCad 10.0.6's own connectivity check.

No new errors

No KiCad DRC errors beyond the input board's own. Every input starts with zero errors.

Board unchanged

Footprints, pads, nets, outline, layers, design rules and project settings identical to the input.

Task respected

Reroutes keep all other copper exactly. Pour tasks keep zone outlines, nets and priority, and the ground stays connected after a refill.

Every router gets the same rules and the same checker, and every verdict reproduces in plain KiCad DRC. Before the study, every case was shown to have at least one valid board: a router's output, the original human routing, or a small hand repair. Three runs per case are repeats for timing and consistency, not different boards; no router ever split a verdict across its three runs.

Environment

Machine: Intel i9-14900HX, 32 logical CPUs, 30 GB RAM
OS: Ubuntu 24.04, Linux 6.8
Checker: KiCad 10.0.6 (connectivity and DRC)
Scheduling: One run at a time; no CPU pinning
Time limit: None, apart from a 60-minute safety stop for hung runs
Threading: Each router's own default, so compute is not equal. Average busy cores in the ranking table
Run days: 273 runs on 9 October, 117 reruns on 12 October 2026
Settings: Tuned per router and task on separate boards, then frozen before scoring

Routers and settings

Router	Version	Reads the board as	Key settings	Threading
drandyhaas/KiCadRoutingTools	0.23.0 (9399bfb)	KiCad file directly	0.05 mm grid (0.1 mm for pours); smoothing off; keeps input copper	Runtime default (no override)
TraceMaker	0.9.0 CPU build (24c8477)	KiCad file directly	No GPU, no learned state between runs; via cost 3 mm; soft zones for pours	8-thread portfolio by default
fastroute	0.1.13	Specctra DSN export	Enhanced default mode; neckdown off; one start; up to 100 passes; fanout off	Parallel autorouter on all cores
fab_router	69cf598	Specctra DSN export	Up to 100 passes; fanout off; no neck-down; strict DRC on for pours	Synchronous; no worker override
Freerouting	2.5.0	Specctra DSN export	Up to 100 passes; neckdown and fanout off; analytics off; JDK 25	Optimizer pool of 31 threads

Routers that read a Specctra DSN export cannot see KiCad's thermal-relief or zone-priority rules; pours are refilled natively after routing. Coupled differential pairs were not exercised: the only pair case had no provable valid board and was removed.

Boards

All boards are public, openly licensed 2-layer KiCad projects, used unchanged apart from removing copper for each task and, for power-width cases, declaring the required widths.

Audits

Stale pour shapes

Our checker judged reroutes against old pour shapes. Fixed; 30 runs that should have passed now pass.

A rule stricter than KiCad

An extra width rule of ours failed fastroute 27 times. Dropped, so KiCad's own rules decide.

Setup fixes

A TraceMaker pre-check crash and duplicate zone IDs fixed; two router settings corrected and rerun.

Final check

A second audit reproduced all 237 passes and 153 fails in plain KiCad DRC, and the scores recompute exactly.

Limits

11 boards, all 2-layer. Results may differ on other boards or with other settings.
No differential-pair result.
Electrical, signal-integrity and manufacturing quality are not measured.
Runs come from two days, so close speed differences are descriptive only.
Reroute vias and track length include the copper that was kept.
fab_router can place layer changes without vias; that is its own defect.