Autorouter benchmarks, scored by KiCad

Measured 2026-10-14/16, KiCad 10.0.4 as referee, one machine for every tool. Koppar’s time is the whole product path, everything a user’s own board goes through. Freerouting writes its board only when it finishes, so a run past the cap has no board to score. tscircuit, KiCad Routing Tools, TraceMaker and Zap! boards are scored exactly as returned, with any pours they drew refilled. Zap! is a hosted service: its time is the job’s run time on the service, not its queue, and it refused one board whose human design has a footprint outside the outline. Times are when a tool reached the board it returned, beside how long it ran: a router that keeps searching after its best board shows both. Koppar’s cell shows two modes: the first line is its default, which keeps every via out of pin pads, so the board needs no special fab process; the second is its filled-via mode, which lets vias sit in pin pads, for a board ordered with filled and capped vias. Each cell counts the vias a tool left in pin pads.

The referee

One pinned KiCad 10 install runs kicad-cli DRC and reads the ratsnest. No tool scores its own output; self-reported completion is recorded but not published.

The boards

KiCad demos, Raspberry Pi reference designs, OLIMEX and Antmicro carriers, each under its upstream licence. Each router gets the placement only: tracks, vias and copper pours are removed, and every net, ground included, is the router’s to connect. The design rules are the benchmark’s: one common floor, published per board and applied to every tool and to the referee, so the table measures routing under that floor, not each design’s own constraints.

The metric

KiCad airwires. Net-level completion reads high on a board with a stranded pad or a severed pour; the ratsnest does not.

One configuration

The same configuration for every board, as for a customer board. No per-board flags for any tool.

• No human parity. Koppar routes every board here, but the human layout is still better on vias, track length and plane integrity, and TraceMaker drills fewer vias than Koppar on every 4-layer board. Completion and DRC do not score planes: a 4-layer board routed with no ground plane can reach 100% here.
• Each tool was run as shipped, with its defaults and the board’s rules. A tool’s author may know settings that do better; the harnesses are in the open so anyone can rerun them.
• Completion is not quality. Rail sizing, decoupling distance and return paths are not scored yet.
• Eight boards is a small sample. 25 more public open-hardware boards, 2 to 8 layers, are pinned in the benchmark repository; the run on them is coming.