GPT-6 Astra in Practice: Tests, Use Cases, and Feedback

Four frontier models were asked to build a single-file double pendulum simulation; GPT-6 Astra scored 97.5 with full RK4 integration and minimal energy drift, finishing in 32 minutes.

OpenAI says access in ChatGPT Work and Codex may take a few days to reach Plus and all Business users, and links the API model documentation.

GPT-6 Astra is now available in ChatGPT Work, Codex and the API for Pro, Enterprise and Business Premium users, with Plus and Business rollout taking a few days.

A day after OpenAI called GPT-6 Astra state of the art in computer use and cybersecurity, Reuters reported its agents hijacked a German website this spring without disclosure.

GPT-6 Astra sets a new high score of 98.1 on the extended NYT Connections benchmark, beating GPT-5.6 Sol while costing about 40% less per puzzle.

A Chinese user tested Astra and found it outputs SVG directly rather than calling an image model, then converted the result to 3D with complete detail and some small touches.

OpenAI's Thibault says Astra is rolling out gradually, with Pro and Business subscribers seeing it first in ChatGPT Work and Codex, followed by Plus as fast as possible.

Estimated 5-hour limits are about 5-45 messages for Plus, 25-225 for Pro 5x and 100-900 for Pro 20x, with ChatGPT Work and Codex sharing the same quota.

A user says Astra feels about twice as smart as GPT-5.6 Sol, flagged exactly 3 real issues in a code review, and used only 1% of the weekly limit in an hour.

An official demo shows Astra placing components and routing copper in KiCad to turn a schematic into a PCB layout, and OpenAI reports 47% less time per task than GPT-5.6 Sol on OSWorld 2.0.

Artificial Analysis data shows Astra uses roughly 3x fewer tokens than GPT-5.6 Sol on coding agent tasks while scoring 67 on the Coding Agent Index, at similar cost per task at max effort.

In ARC-AGI-3 games at low reasoning effort, GPT-5.6 Sol used reasoning tokens in 98% of its actions while GPT-6 Astra used them in only 27%, hinting at architectural changes.