The company's own table shows gains on coding, engineering and legal tests. On Terminal-Bench 4.0 it reports 38%, where an independent evaluator measured 26%.
Benchmark
EEBench
Whether a model can design a working electronic circuit. Version 1 covers 13 analog and digital design tasks and scores both electrical correctness and practical engineering trade-offs such as component cost and availability.
- Maintainer
- atopile
- Score
- %
- Last checked
- September 5, 2026
Leaderboard
No verified results recorded yet.
Methodology
Grading is deterministic rather than human or model-judged. The harness builds each submitted design, constructs the circuit graph and bill of materials, then runs SPICE simulations and design rule checks against the task's requirements. The maintainers compare it to giving a coding agent a compiler and a test suite, except that the tests measure voltages and component behaviour.
Important limitations
Simulation is not manufacture. The benchmark does not test whether a model can lay out a board, get it produced or bring up working hardware, and the maintainers say physical validation is planned for a later version. It is also built and funded by atopile, a vendor in the same field, and the task set is small at thirteen items.
Related news
A benchmark grades AI circuit designs with SPICE, and the best model manages 61.6%
The tasks are graded by building the design, extracting a bill of materials and running SPICE against the requirements. Nothing gets marked correct by opinion.