# NVIDIA describes NeMo Gym evaluation for a chip-design agent

> In a development benchmark reported by NVIDIA, Siemens EDA cut agent turns from 155 to 53 while raising the composite verification score from about 55% to 95%.

- Canonical URL: https://newsverum.com/en/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026
- Publisher: Newsverum
- Section: Technology
- Published: 2026-10-15T19:00:45+00:00
- Updated: 2026-10-15T19:00:45+00:00
- Event / data period: 2026-10-14
- Other editions: [Türkçe](https://newsverum.com/tr/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md), [العربية](https://newsverum.com/ar/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md), [Français](https://newsverum.com/fr/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md), [Español](https://newsverum.com/es/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md), [Deutsch](https://newsverum.com/de/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md), [Português](https://newsverum.com/pt/news/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026.md)

NVIDIA said Siemens EDA used NeMo Gym to test Liberty files generated by its Fuse EDA AI agent during standard-cell characterization. A verifier checked cell and operating-condition coverage, file validity, and whether timing and power values met tolerance. The team reviewed failures in those checks and iteratively changed the agent's skills and Model Context Protocol tool descriptions.

In the development benchmark NVIDIA reported, the composite verification score rose from about 55% to 95%, agent turns fell from 155 to 53 and output tokens declined from about 81,000 to 29,000. The setup covered 229 cells and 36 process conditions from the open GF180MCU process design kit, while agent-tuning runs evaluated two cells; the team conducted about 25 iterations. These are NVIDIA and Siemens' results on representative tasks, not a general measure of chip-design productivity.

## Data

### Measured changes for the Siemens EDA agent

Approximate score and workflow counts reported by NVIDIA for its standard-cell-characterization benchmark.

Methodology: Before-and-after values reported in NVIDIA's 14 October 2026 Siemens EDA case study after refining agent skills and MCP tool descriptions. The benchmark used 229 standard cells and 36 PVT corners from the GF180MCU PDK; agent-tuning rollouts evaluated two cells. Results are vendor-reported development-benchmark figures for representative tasks.

| label | value | unit | period | basis | value_qualifier |
|---|---|---|---|---|---|
| Composite verification score, before | 55 | percent | 2026-10-14 | observed | approximate |
| Composite verification score, after | 95 | percent | 2026-10-14 | observed | approximate |
| Agent turns, before | 155 | count | 2026-10-14 | observed |  |
| Agent turns, after | 53 | count | 2026-10-14 | observed |  |
| Output tokens, before | 81000 | count | 2026-10-14 | observed | approximate |
| Output tokens, after | 29000 | count | 2026-10-14 | observed | approximate |

Sources: <https://developer.nvidia.com/blog/optimizing-agent-reliability-and-tool-calling-efficiency-for-chip-design-using-nvidia-nemo-gym/> · Download data: [JSON](https://newsverum.com/api/v1/articles/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026/data/siemens-eda-agent-evaluation?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/nvidia-nemo-gym-improves-chip-design-agent-evaluation-october-2026/data/siemens-eda-agent-evaluation?lang=en&format=csv)

## Sources

- [NVIDIA Technical Blog, Siemens EDA case study](<https://developer.nvidia.com/blog/optimizing-agent-reliability-and-tool-calling-efficiency-for-chip-design-using-nvidia-nemo-gym/>) (Source checked: 2026-10-15; Published: 2026-10-14)

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)