What changed
OpenAI says model-proposed changes reduced standard-cell area by about 13% in Jalapeño’s vector unit and 10% in its matrix unit, while AMD reports up to 14% improvement over optimized GPU IP baselines and single-digit gains on next-generation CPU blocks.
Why it matters
The report moves agentic chip design from toy benchmarks toward silicon-proven and manufacturing-bound changes, while keeping expert verification and EDA equivalence checks in the loop.
What to do
Start with an open RTL block and a fixed PPA baseline; require formal equivalence, synthesis, timing and expert review for every proposal, and report failed or reverted variants alongside gains.
Caveat
Results are reported by OpenAI and partners, not independently reproduced, and detailed production designs are not public. Gains vary by block, baseline, toolchain and verification budget.