atopile · 2026-09-04 · major
EEBench — atopile's benchmark scores frontier models on circuit design
EEBench is a benchmark from atopile that grades AI models on 13 circuit-design tasks with SPICE simulation instead of human judgement. Claude Opus 5 leads the first leaderboard at 61.6%, ahead of Grok 4.6 at 57.1%.
A simulation-graded benchmark that asks whether AI models can design circuit boards that actually work.
Quick facts
| Maker | atopile |
|---|---|
| Version | EEBench V1 |
| Tasks | 13 analog and digital design tasks |
| Grading | SPICE simulation and design checks at worst-case tolerance corners |
| Score formula | 0.65 technical + 0.35 cost efficiency |
| Design input | atopile code, not prose or CAD clicks |
| Submissions | By request; self-serve planned |
Benchmarks
| Claude Opus 5 | 61.6% | |
|---|---|---|
| Grok 4.6 | 57.1% | |
| Claude Fable 5.1 | 56.4% | |
| Claude Fable 5 | 54.3% | |
| Claude Opus 4.8 Max | 51.4% | |
| GPT-5.5 | 42.3% | |
| GPT-5.6 Sol | 39.4% |
What is it?
EEBench V1 puts frontier models through 13 circuit-design tasks and grades the answer with physics, not a human vote. The tasks span analog and digital design and follow the loop an electrical engineer really works through: write the requirements, produce a design, then verify it. atopile, the team behind the code-first circuit design tool of the same name, built and funded the benchmark.
How does it work?
Models write atopile design code rather than clicking through a CAD tool, so the grader can read components, connections and electrical constraints directly. Each submission is built into a circuit graph and a bill of materials, then run through SPICE simulation and design checks that measure gain, transient response, thresholds, ripple and tolerance margin at worst-case component corners instead of nominal values. A final score weights technical performance at 0.65 and cost efficiency against a reference bill of materials at 0.35.
Why does it matter?
Hardware has had no SWE-bench of its own, so a claim that a model can route a board was hard to check. The EEBench leaderboard turns it into a number, and the spread is wide: Claude Opus 5 leads at 61.6% while GPT-5.6 Sol sits at 39.4%. atopile's own reading is that for a useful and growing set of circuit problems the answer is already yes, with a lot of room left above 61.6%.
Who is it for?
hardware engineers and model evaluators
Frequently asked questions
- Which model tops the EEBench leaderboard?
- Claude Opus 5 leads EEBench V1 with 61.6%, followed by Grok 4.6 at 57.1%, Claude Fable 5.1 at 56.4%, Claude Fable 5 at 54.3% and Claude Opus 4.8 Max at 51.4%. GPT-5.5 scores 42.3% and GPT-5.6 Sol 39.4%. These figures are the leaderboard snapshot atopile published for 1 September 2026.
- Can I submit my own model to EEBench?
- Not yet on a self-serve basis. atopile says public self-serve submissions are coming later, and asks teams to contact them in the meantime to have a public or internal model evaluated on the held-out benchmark and receive a clean score report. Submissions take the form of atopile design code.
- Why does EEBench score cost as well as correctness?
- A circuit that meets every requirement with expensive parts is not a good design, so EEBench gives 35% of the score to cost efficiency. It compares the model's bill of materials against a reference design at distributor pricing for quantity 100, and the remaining 65% comes from the simulated technical measurements.
- Does a top EEBench score mean AI can replace a hardware engineer?
- No. atopile writes that for a useful and growing set of circuit problems models can already design working circuits, while acknowledging substantial progress is still needed. The leading score on EEBench V1 is 61.6%, so more than a third of the measured requirements are still missed by the best model tested.
Try it
https://eebench.org/ — full leaderboard, per-task scores, cost and time per task