OpenAI releases first performance data for Jalapeño AI chip

Investing.com -- OpenAI has published the initial performance results for Jalapeño, its first custom inference chip, revealing notable gains in both speed and efficiency. Shattering the traditional trade-off between throughput and latency, Jalapeño delivers massive upgrades in power efficiency and processing time compared to today’s leading commercial AI hardware.

The company tested Jalapeño using InferenceX, a public benchmark from SemiAnalysis, comparing it against leading commercially available AI systems. The tests covered three models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T.

Jalapeño delivered 1.5 to 1.9 times more AI work per watt at peak throughput and 1.7 to 3.6 times lower end-to-end latency than comparison systems across all three models. For highly interactive workloads, it delivered 2.1 to 4.1 times higher performance.

On the Kimi K2.5 1T model, the largest public model tested, Jalapeño delivered approximately 1.5 times higher peak performance per watt and 3.4 times lower end-to-end latency than comparison systems. The chip is rated at 700 watts, though measured sustained power remained at or below 550 watts during testing.

OpenAI stated that the chip’s performance advantage widened further on its internal frontier models, suggesting the architecture becomes more effective as workloads grow larger.

The development process from initial design to tapeout took nine months. OpenAI used its own AI models to help design and optimize the chip, with earlier generations assisting the design phase and newer models accelerating optimization and programming.

Using Codex with GPT-Astra, the team brought three open-weight models to high performance within two months. For selected GPT-OSS attention and mixture-of-experts blocks, AI-generated implementations ran 1.5 to 1.8 times faster than existing human-written implementations.

OpenAI plans to begin deploying Jalapeño within its compute infrastructure by the end of the year. The company stated that a second generation is in development and a third generation is taking shape.

The company said it will continue to deploy accelerators from Nvidia and other partners for both training and inference workloads alongside Jalapeño.

You May Also Be Interested In

• Raymond James upgrades AMD, sees $201 billion server CPU market by 2030
• Truflation sees July core PCE at 0.2% MoM, says Fed won’t hike this year
• XPeng stock falls on Q2 results miss; robotics arm raises $900 mln