OpenAI Unveils GPT-6 Astra and Plans Year-End Deployment of Proprietary Inference Chip 'Jalapeño'

OpenAI unveiled GPT-6 Astra, stating that it has achieved state-of-the-art performance in computer use, software engineering, and cybersecurity. With 1 billion weekly active users (WAU) and 2.5 million enterprise customers secured, OpenAI is accelerating its strategy to translate research progress into consumer and enterprise benefits.

Proprietary Inference Chip 'Jalapeño' and Cost Reduction

OpenAI unveiled its first self-developed inference chip, Jalapeño. InferenceX test results showed that peak token throughput per watt is 1.5 to 1.9 times higher than commercial systems, and end-to-end latency is 1.7 to 3.6 times lower. This chip is scheduled for deployment by year-end alongside partner accelerators such as NVIDIA and AMD. Additionally, serving software improvements for GPT-5.6 Sol have reduced end-to-end serving costs by 20% and improved token-generation efficiency by more than 15%.

Maximizing Research and Work Efficiency

OpenAI's internal research organization achieved a productivity metric utilizing 3.1 agent-workdays of effort per human workday. The structure involves delegating complex tasks to agents while humans focus on prioritization and result judgment. These agentic products blur the boundaries between personal and professional life, creating a virtuous cycle where ChatGPT experiences at home lead to the adoption of work tools.

Key Customer Cases

• Boston Children’s Hospital: AI-assisted research yielded over 40 diagnoses in previously unresolved rare disease cases.
• Balyasny Asset Management: Reduced macroeconomic scenario analysis time from two days to approximately 30 minutes.
• Circles: Autonomously resolved 65% of customer service interactions.
• Cars24: Supports vehicle comparison and booking by processing over 1 million minutes of conversations per month.