OpenAI has released performance data for its custom inference chip, codenamed Jalapeño. The chip reportedly achieved higher peak throughput per kilowatt and lower token latency than existing commercial systems when tested on the InferenceX benchmark using the GPT-OSS 120B model. This marks a significant step for OpenAI in developing its own specialized hardware for AI inference. AI
IMPACT Demonstrates OpenAI's move towards custom hardware, potentially improving inference efficiency and reducing reliance on third-party providers.
RANK_REASON Frontier-lab announcement of custom inference chip performance. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on X — Greg Brockman (OpenAI) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →