OpenAI launches GPT-6 Astra and custom silicon
OpenAI announced GPT-6 Astra alongside Jalapeño, its first custom inference chip, in a strategy post that pairs a frontier model launch with hardware and serving-cost claims but publishes no benchmarks.
In a post titled "The Work Now Within Reach," OpenAI introduced GPT‑6 Astra, which it calls "the world's most intelligent and aligned model" and state-of-the-art in computer use, browsing, software engineering, cybersecurity, science, and professional work. No evals, benchmark numbers, or comparison models accompany the claim.
The company also detailed Jalapeño, its first custom inference chip. In InferenceX tests across three public models, OpenAI says Jalapeño delivered 1.5–1.9× the peak token throughput per watt of the commercial systems tested — normalized by rated chip power — with end-to-end latency 1.7–3.6× lower. Deployment is planned to begin by year-end, alongside accelerators from NVIDIA, AMD and other partners. The comparison systems are not named.
On serving economics, OpenAI says GPT‑5.6 Sol helped improve its production serving software, cutting end-to-end serving costs by 20%, with further work raising token-generation efficiency by more than 15%.
The post also puts scale figures on the record: more than one billion weekly active users and 2.5 million businesses across ChatGPT, ChatGPT Work, Codex and the API. OpenAI says a study of individual ChatGPT subscribers found daily message volume roughly 50% higher six months after signup than in the first month, across roughly twice as many distinct tasks. Internally, it says its research organization now uses 3.1 agent-workdays of effort per workday of human labor.
The piece is investor-facing framing — reach, full-stack compute, capital discipline — rather than a technical launch. Pricing, availability and model specifications were not disclosed.