← All benchmarks

Build-A-Chip-Bench (20-task index run)

Agents

ChipAgents' 20-task Build-A-Chip-Bench run shared for Fireworks' Specialized Intelligence Index: each task takes a workload specification and pinned interface through architecture, RTL, synthesis, place-and-route and signoff to a GDS layout with OpenCode, open-source tools and an open-source PDK in 72 hours. A design scores its PPA against the task's constraints (0-100%) only if it passes hidden functional verification and the clean-GDS gate, and zero otherwise; the score is the mean over the 20 tasks. It is kept as a component beside the write-up's earlier Table 1 run, whose task count is unpublished and whose shared models score differently.

Top models (higher is better)