DeepPCB vs Quilter: 3 Open-Source Boards, Public Tools, Real Numbers
Both tools publicly available. Three open-source boards. Identical conditions. No manual intervention on either side.
In March 2026, Quilter published an SEO article titled “The 2026 Guide to Autonomous PCB Design: Quilter vs. DeepPCB vs. Flux.ai.” In it, they describe DeepPCB as suited for “simpler constraints” and “earlier-stage programs,” and cite internal benchmarks showing Quilter ahead on placement success and routing time.
We wanted to see what the numbers look like when anyone can verify the results.
We took three open-source KiCad boards, ran them through the publicly available versions of both tools (deeppcb.ai and quilter.ai), and recorded everything. No manual intervention. No parameter tuning. No placement changes. Just upload and route.
Any engineer can reproduce this test. The boards are on GitHub. Both tools have public access.
The Boards
All three are open-source KiCad designs, chosen for varying complexity:
- STRF: github.com/pms67/STRF-Kicad. 98 airwires. Simpler evaluation board.
- PocketBeagle: github.com/beagleboard/pocketbeagle. 290 airwires. Mid-complexity single-board computer.
- BeagleConnect Freedom: git.beagleboard.org/beagleconnect/freedom. 414 airwires. Complex mixed-signal IoT board.
The Results
STRF (98 airwires)
| Metric | DeepPCB | Quilter |
|---|---|---|
| Completion | 99% | 94% |
| Vias | 29 | 58 |
PocketBeagle (290 airwires)
| Metric | DeepPCB | Quilter |
|---|---|---|
| Completion | 96% | 82% |
| Vias | 135 | 163 |
BeagleConnect Freedom (414 airwires)
| Metric | DeepPCB | Quilter |
|---|---|---|
| Completion | 97% | 87% |
| Vias | 191 | 235 |
What the Numbers Say
| Metric | DeepPCB | Quilter |
|---|---|---|
| Average completion | 97.3% | 87.7% |
| Via count (relative) | Baseline | +44% more |
| Time to first result | 1-2 minutes | 15-20 minutes |
| Constraint extraction | Automatic from file | Manual entry |
| Output formats | 6 | 2 |
Both tools are doing real work. Quilter routes boards. The question is what the output looks like on identical inputs.
On these three boards, DeepPCB completed 10 percentage points more connections on average, used 44% fewer vias, and produced its first result roughly 10x faster.
Why This Matters
Via count is not a vanity metric. Every via is a drilled hole, an impedance discontinuity, and a potential failure point. On the STRF board, Quilter placed 58 vias where DeepPCB placed 29. At production volume, that difference affects yield, cost, and reliability.
Completion rate determines how much manual work remains. At 87% on a 414-airwire board, the engineer finishes 54 nets by hand. At 97%, it’s 12 nets. That’s the difference between hours of cleanup and a quick review.
A Note on Quilter’s Published Comparison
Quilter’s March 2026 blog post uses “Quilter internal benchmarks” as its data source. The methodology and boards are not published. We can’t reproduce their numbers.
The three boards above are public. The tools are public. Any engineer with 10 minutes can verify these results. We think that’s a better standard for a benchmark.
We are not claiming DeepPCB is better at everything. Both tools have different strengths and different architectures. Quilter offers a standalone autonomous design flow. DeepPCB is a routing engine that integrates into existing EDA workflows via API. Different products for different use cases.
The data on these three boards is what it is.
Reproduce It Yourself
The best benchmark is always your own board. But if you want to start with these:
- Download any of the three boards from GitHub
- Upload to deeppcb.ai and quilter.ai
- Compare completion rate, via count, and time to result
Both tools have free or trial access. The data speaks for itself.
Board sources: STRF, PocketBeagle, BeagleConnect Freedom
