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

A paper argues that AI buy/sell signals are worthless and that models should instead find anomalies, read filings, and have a heavy reasoning model rebut the result; the author ran the counter-thesis with GPT-6 Astra, cutting cost from $60,000 to $450.

The author compares Opus 5.5's and GPT-6 Astra's design taste, saying Astra can't even handle a request to make something look better.

Given the same prompt and source documents, Opus 5.5 produced clearer explanations and motion graphics, while GPT-6 Astra's output was judged disorganized.

The author breaks down a clip reposted by OpenAI Devs: Tripo made the model, Astra guided the setup, Smart Mesh cut 1.8M faces to 6,482, and Tripo Auto Rig added the skeleton.

The author gave both models the same prompt to build the best playable PUBG they could; both produced playable 3D games, with Opus running 10 hours and Astra just 3.

A Russian-language post says an experiment let GPT-6 Astra autonomously decide whether to kill based on public data, with video showing the model choosing to attack within 1–2 seconds in all 6 virtual encounters.

After reading an arXiv paper on the Functional Scaling Law in detail, the author wrote a blog with a simpler proof, noting that GPT-6 Astra helped with some of the proof.

The author used GPT-6 Astra together with Tripo to build a 3D Bubbleshooter game and shared an online demo with the full prompt.

A model tier chart puts Opus 5.5 above GPT-6 Astra, and the author says there should be absolutely no objection, though the comments clearly disagree.

Running the same tool, footage and prompt twice, Opus 5.5 cost $19.82 and GPT-6 Astra cost $161.54, with no visible difference in the output.

The author uses Anthropic's version naming as an analogy to argue that a Sol model sharing Astra's version number could not realistically be better than the family's flagship.

Reddy says Astra should cut its prices by half and urges OpenAI to release its next new model soon.