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

A user shares a purely subjective ranking of 14 models they have used, with Claude Opus 5.5 first and GPT-6 Astra third.

Given only a scan of a 1969 journal page, Astra transcribed about 1,300 hand-drawn symbols, completed the key and solved an 1809 coded letter, also correcting the original letter's date.

A Persian version of the same Napoleonic general's coded-letter case, stressing that Astra did all the decryption work starting from a single image.

A short image-free comment in which the author only says what Astra managed to do here was impressive, without naming the task.

The author used GPT-6 Astra to generate a playable F1 game set on a Kuala Lumpur circuit, then had Claude Opus 5.5 turn a screen recording into a trailer with start lights and a soundtrack.

At its September 29 DevDay, OpenAI launched always-on Dot agents, GPT-6.1 Sol, Codex background tasks and more, but declined to release its flagship GPT-6.1 Astra on safety grounds.

The author says the entire animated short was generated with code, with no video model or stock footage, and that this was only a first test.

The brief covers Google's Gemini 4 Argon entering trusted testing, Anthropic and Grok updates, and shareable profiles going live in ChatGPT.

Given a scan of an encrypted letter sent to one of Napoleon's generals 217 years ago, Astra transcribed about 1,300 cipher symbols, found a partial historical key, built a solver for the rest, decoded the letter, and noticed the original date was wrong.

The poster says they upgraded their subscription to the Pro plan specifically to use dots, which runs on GPT-6 Astra, and included a video demo.

A Japanese post says GPT-6 Astra spent about six hours cracking a Napoleonic army cipher unread for 217 years, working through 1,300 cipher symbols.

On the same prompt, Opus's cherry blossom tree looks fuller and prettier while Astra's has a more pronounced jelly texture and costs nearly $6 less.