OpenAI presented test results showing its Jalapeño processor outperformed Nvidia processors.
The ChatGPT owner said the chip led on work per unit of power and response speed when measured against Nvidia's GB300 on the public benchmarking system the team used.
A separate report said Jalapeño outpaced GB200- and GB300-based systems on energy efficiency and latency in inference tests.
The design, co-developed with Broadcom, was first shown in June and OpenAI has been running samples in its labs as it prepares the processor for inference workloads.
The effort to build custom accelerators is part of OpenAI's broader push to reduce dependence on external GPU suppliers and to lower power and operating cost for its expanding frontier models.
The claim arrives while the company faces regulatory pressure, including an Alabama probe opened after a model-related breach at another AI firm.
OpenAI has said the Jalapeño design will begin entering commercial use by the end of 2026, with broader deployment expected to ramp through 2027.
