Summary
OpenAI says Jalapeño delivered higher throughput, lower latency and better energy efficiency than comparison systems when serving GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T. The results are vendor benchmarks for inference hardware, not training, and OpenAI says limited systems will be deployed this year.
Why it matters
The results show frontier-model providers are increasingly designing specialised inference hardware to control cost, capacity and latency, although OpenAI remains dependent on Nvidia for training.
Related coverage
- Nvidia doubles entry-level edge compute with Jetson Orin Nano 2
- Perplexity launches a local-first Computer agent for Nvidia DGX Spark
- Generalist raises another $200 million for robot foundation models