Atlas

Benchmarks

← All benchmarks

Build-A-Chip-Bench

Agents · 2026-10-14

ChipAgents' end-to-end chip-design agent benchmark, developed with contributions from NVIDIA: each task takes a workload specification through architecture, RTL, synthesis, place-and-route and signoff to a GDS layout with open-source tools and an open-source PDK under a fixed time budget. A design scores its PPA against the task's constraints (0-100%) only if it passes a hidden-VIP functional gate and a clean-GDS gate (routable, hold-clean, DRC-clean), and zero otherwise; the score is the mean over tasks. The task count is not published.

Top models (higher is better)

ModelScore
Claude Fable 5.116.6
GPT-6 Astra8.7
Gemini 3.8 Flash0.8
Kimi K30.2
GLM-5.30.0
Loading Atlas data…