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

A prompt to paste after planning and before an agent writes code, pushing it to check that the approach is the most reliable, not overengineered, and free of drifting assumptions.

On the same messy production-style repo with a vague prompt, Haiku 5.5 and GPT-6 Astra both found 8 of 9 hidden defects, but Astra touched 68 files while Haiku changed only 34.

A Spanish-language post jokes about Adobe's CEO watching a GPT-6 Astra recreation of Photoshop published for free, with nothing he can do since no law was technically broken.

A test found Astra precisely follows a requested work duration only for tasks over 2 hours, finishing exactly at 72 hours when asked, while Fable 5.1 quit after 3.

Rasmic, who built a custom payment system for his church that handles a lot of money, had GPT-6 Astra audit it and found one security issue that would have been detrimental, urging anyone with real users and money involved to run an audit.

A developer rewrote the Rust PNG compression tool oxipng in TypeScript, reaching about 60% of the original's performance, with the work done almost entirely by GPT-6 Astra / 6.1 Sol.

A Max subscriber says GPT-6 Astra and Sol use up quota faster with limited improvement, that Astra is inconsistent, and that only Computer Use and image generation are worth using.

The author poses a hypothetical: both models have unlimited usage, but you can only use one for the next year, updates included, and asks which people would pick and why.

A user says they were excited when Astra launched but that things have clearly gotten worse over recent weeks, and urges OpenAI to take users' concerns seriously.

Given the same four-word prompt "SUBWAY SURFER!!!!", both models generated a playable 3D runner in real time, which the author uses to compare taste and ability to fill in gaps beyond code quality.

TrendSpider shared a screenshot of a Sunday backtest of $TSLA run with GPT-6 Astra, showing how it is used for quantitative analysis.

An Arabic post introduces an open-source tool that uses an existing ChatGPT subscription to read and edit project files, run commands and tests, and run multiple agents in parallel, though ChatGPT subscription limits still apply.