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

A post lists recent OpenAI grievances: GPT-6.1 Astra shelved, Sol slow, Codex limits hard to rely on, Pro cut from 20x to 10x, and a new $500 tier.

The author argues Astra follows instructions so strongly that vague or conflicting AGENTS.md wording backfires, and summarizes how to write one that fits it.

An observation says ChatGPT on the web now spawns subagents via GPT-6 Astra Pro even for simple tasks, with speculation that Chat and Work will merge soon.

A tester paired Astra with a self-built crawler agent that found a missed invoice across Gmail and Slack and drafted it for approval, which the author calls a 24/7 chief of staff.

A critic argues the three new models are really one model, with Sol a slowed-down Astra whose speed is added back through Ultrafast.

A critical post says Astra put Codex ahead three weeks ago, but usage was cut around DevDay and the $200 Pro plan now includes less, while Anthropic raised limits and shipped Opus 5.5.

The author compares Opus 5.5 and GPT-6 Astra on frontend design taste, with images showing each model's output.

The author argues that new model releases are always overhyped: Opus 5.5 is very capable but has to be pushed repeatedly, and only starts working once told to "just build it".

Tibo announced that OpenAI will ship a new feature every day for 28 days for Codex/Work users, with a full usage reset for everyone if a day is missed; the first two days brought Astra/Sol at 50% faster and free auto-review.

Someone mocked OpenAI's new model naming as a "Day-2.1" series, stringing together names like Astra, Sol and Luna as a joke.

A daily roundup says GPT-6 Astra and GPT-6.1 Sol are now about 50% faster by default, and that OpenAI also shipped TextGrain, an invisible text watermark.

The author says GPT-6 Astra, Opus 5.5 and Jev replaced five job roles, and by the second month the bill was down to electricity and a small amount of API usage.