Briefing
For each design the harness builds the atopile project, generates the circuit graph and BOM, and runs SPICE and design checks. Cost counts toward the score only if the circuit works. In the 1 September results, Grok 4.6 followed at 57.1%, then Claude Fable 5.1 at 56.4%, Claude Fable 5 at 54.3% and Claude Opus 4.8 Max at 51.4%.
The capacitor failure was in a hold-up design that needed 545 uF. At 4.7 V bias the nominal 22 uF part gave about 11.4 uF, and the rail fell below 3 V after 0.85 ms instead of holding for 20 ms. The atopile team built and funded the benchmark, and all 13 tasks are written in its language.
Written by the Boards & Bots curator