OpenAI says its models are improving Jalapeño and AMD chip designs
OpenAI says its models have delivered silicon-proven improvements to its Jalapeño inference chip and up to 14% gains against optimized baselines on AMD hardware. The company also reported stronger results from newer models on open-source RISC-V processor design tasks.
- OpenAI estimates area reductions of about 13% in Jalapeño’s warp engine and 10% in its matrix unit.
- The company reports up to 14% PPA improvement on AMD flagship GPU IP.
- GPT‑6.2 Astra scored highest on both tested RISC-V cores at comparable budgets.
- The RISC-V evaluation included 9,881 scored runs.
- OpenAI is partnering with Synopsys on GPT‑Synopsys.
Model-assisted improvements to Jalapeño
OpenAI began using GPT‑4o to help design Jalapeño in early 2025. According to the company, the model found improvements in the chip’s warp engine and matrix unit by exploring alternative logic implementations, pipeline organization, hardware sharing and reduced intermediate storage.
One redesigned multiplier in the matrix unit used about 15% less cell area in synthesis and reduced latency from two clock cycles to one. A separate adder redesign reduced cell area by about 15% while keeping its two-cycle latency. OpenAI estimates that model-proposed changes reduced standard-cell area by about 13% in the warp engine and 10% in the matrix unit. The company says those optimizations were proven in silicon during Jalapeño’s silicon bring-up.
AMD reports gains on production designs
OpenAI says it has partnered with AMD on power, performance and area, or PPA, improvements and achieved up to 14% improvement relative to heavily optimized baselines on AMD’s flagship GPU IP. Its model rewrote SystemVerilog RTL, the code that describes a chip’s circuits, and used Synopsys tools to verify that the changes preserved the blocks’ behavior.
According to OpenAI, AMD has included some of the changes in a recent design sent for manufacturing and added others to production design files for future versions. AMD also verified single-digit percentage PPA gains across key blocks of its upcoming next-generation flagship CPU using designs written by OpenAI’s model. OpenAI says the work represents roughly a year of progress achieved in less than 5 months, based on its comparison with prior chip design progress.
Newer models lead RISC-V evaluations
To track progress between model generations, OpenAI tested models on PicoRV32 and Ibex, two open-source 32-bit RISC-V cores. The models were asked both to improve existing hardware descriptions and to design a core from written requirements. The evaluation measured estimated instructions per second, while rejecting designs that failed correctness checks or used more than 16 times the original core’s area.
At a comparable budget of about 112,000 tokens per run, GPT‑6.2 Astra had the highest average score on both cores. GPT‑6.1 Sol also outscored GPT‑5.6 Sol on both. When designing from written requirements, all five models scored higher at about 112,000 tokens than when optimizing the existing designs. OpenAI says the full evaluation included 9,881 scored runs.
The company cautions that specification-based designs could omit features such as performance counters and additional instructions. It says tighter area limits, required features and timing checks after physical synthesis would make the evaluation more representative of production chip design.
OpenAI and Synopsys plan an EDA-focused model
OpenAI says it is partnering with Synopsys to train GPT‑Synopsys, a model intended to use Synopsys electronic design automation tools in the way expert chip designers do. The partnership is meant to connect model development with trusted evaluators and broaden the design challenges covered by the company’s tests.
For users of chip design tools and hardware companies, the announcement points to a role for AI in both optimizing established production designs and exploring processor architectures. The reported results remain OpenAI’s claims, while the work with AMD and Synopsys indicates an effort to test those capabilities against commercial designs and tools.
OpenAI: Advancing chip design with OpenAI models


