# Siemens chip-design agent improves verification score in Nvidia evaluation

_Published Wednesday, October 14, 2026 at 12:26 PM EDT · AI, Infrastructure · Latest · Tier 2 — Notable_

![Siemens chip-design agent improves verification score in Nvidia evaluation — Primary](https://developer-blogs.nvidia.com/wp-content/uploads/2026/10/agentic-workflow.webp)

Siemens improved its chip-characterization agent's mean composite verification score from approximately 55% to 95% by refining instructions and tool descriptions, Nvidia reported. The evaluation checked final files containing timing and power models against reference results, catching missing cells, operating conditions and inaccurate values.

Using NeMo Gym, the team conducted approximately 25 rollouts with Claude Opus 4.8 in Claude Code. Agent turns fell from 155 to 53, and output tokens fell from approximately 81K to 29K. Each run covered only two standard cells across 36 process, voltage and temperature conditions, limiting the demonstrated scope.

## Sources

- [NVIDIA Technical Blog](https://developer.nvidia.com/blog/optimizing-agent-reliability-and-tool-calling-efficiency-for-chip-design-using-nvidia-nemo-gym/)

---
Canonical: https://techandbusiness.org/newswire/EQ2b6zTsGeTv265kcxmwEo
Published: 2026-10-14T16:26:03.367Z
Story chronology: 2026-10-14T16:10:45.000Z
Retrieved: 2026-10-14T19:06:41.707Z
Publisher: Tech & Business (techandbusiness.org)
