OpenAI expands Astra internal testing, launch expected Sept 3
OpenAI expands internal testing of multi-agent model codenamed Astra (mozaik-alpha-fdm); launch expected Sept 3, directly competing with Anthropic Fable 5.1.
On August 31, OpenAI expanded internal testing of its new model codenamed Astra. Registered internally as mozaik-alpha-fdm, the model's core capability is supporting multiple AI agents working in parallel within a unified task framework, sustaining complex long-horizon tasks.
Developer testing shows Astra in Max mode can zero-shot generate 3D isometric maps and interactive web pages, repeatedly validating outputs during generation and maintaining a consistent design language. Core breakthroughs include end-to-end multi-agent orchestration, ultra-long-horizon task persistence, persistent reasoning, and instant self-correction.
Astra is expected to launch around September 3, directly competing with Anthropic Fable 5.1, with OpenAI likely continuing to push API call costs lower.
On the regulatory side, Astra is expected to be among the first frontier models reviewed by federal agencies under the Trump administration's new framework before public release. Executive Order 14409 issued on June 2 has already established the framework.