GPT-6 Astra: what 48 developers actually say.

48 Hacker News opinions from 45 developers, Nov 2025 – Nov 2026. 40 come from people who say they use it. Here is what they praise, what they complain about, and who switched.

48 opinions45 developersNov 2025 – Nov 2026Methodology
Approval
51%
±11.9 at 90% confidence
Last 90 days
51%
↓ 0 pts vs all-time
Rank
#42
of 45 models · CI-adjusted
Opinions
48
40 first-hand
Developers
45
1.1 opinions each
The short version

The case for GPT-6 Astra, in one paragraph.

GPT-6 Astra earns 51% approval across 48 opinions. Developers are divided on it. In the last 90 days it sits at 51%.

Pick it if you

  • Do planning, design documents, and code review more than first-draft coding
  • Are on a $100 Codex plan and want usage that stretches far
  • Want a second model to review another model's code

Look elsewhere if you

  • Need reliable Go or C++ output
  • Need predictable usage limits
  • Need code that lands close to your prompt

Where it wins and loses.

Mentions across 48 opinions. Bars are counts, not votes; a tool can be both praised and criticized on the same topic.

TopicComplaints ← → PraiseMentions
Code quality 18 mentions
810
PraiseComplaint
1 topics shown · sorted by net sentiment · source: 48 opinions
View data table
TopicPraiseComplaintsMentions
Code quality10818

How opinion has moved.

Monthly approval, Sep 2026 – Oct 2026. Bars show opinion volume; low-volume months are still included but faded.

Not enough monthly data for a trend yet.

What developers say.

Summaries are paraphrased. Each links to the original Hacker News comment.

Praise

4 selected
Cost & pricing

Astra in Codex CLI works well for FreeCAD design, though it uses many tokens.

HN comment · Oct 15, 2026Read on HN ↗
Code quality

Found Astra very good at architecture design documents, sometimes beating the author's own solutions.

HN comment · Oct 14, 2026Read on HN ↗
Code quality

Uses Astra for tricky work like planning features and design review.

HN comment · Oct 9, 2026Read on HN ↗
Cost & pricing

Astra on a Codex subscription allows far more usage than a $50 run on a cheap OpenRouter model.

HN comment · Oct 8, 2026Read on HN ↗

Complaints

4 selected
Code quality

Astra code needs tedious, layered fixes to reach good structure and naming.

HN comment · Oct 8, 2026Read on HN ↗
Autonomy

Astra deviates from prompts by taking independent actions, causing data loss while cleaning up a git repository.

HN comment · Oct 4, 2026Read on HN ↗
Cost & pricing

GPT-6 Astra used over twice a weekly budget, stubbed out parts of the implementation, and never finished.

HN comment · Sep 29, 2026Read on HN ↗
Code quality

Astra was poor at writing Go and C++ code, so their team switched back to Claude.

HN comment · Sep 29, 2026Read on HN ↗

Questions developers ask.

Is GPT-6 Astra worth it?
It depends on the job. Approval is 51% across 48 opinions, and 40 come from people who used it. Praise centers on planning, design documents, and code review. Complaints center on heavy cleanup, token burn, and missed prompts.
Is GPT-6 Astra better than Claude Opus 5.5?
Opinions are close. Four people preferred Astra over Claude Opus 5.5, and four preferred Opus 5.5 instead. One team said Astra was poor at Go and C++ and switched back to Claude.
How fast does GPT-6 Astra use up usage limits?
Mixed. Usage limits drew 4 positive and 4 negative mentions. Several users say it drains limits fast: one reported using over twice a weekly budget on a task that never finished, and another said it burns limits that barely move on GPT-5.6 Sol. Some users on a $100 plan say they never ran out.

The data behind this page.

Developers are split on GPT-6 Astra. Approval is 51% across 48 opinions from 45 authors, and 40 of those opinions come from firsthand use. Code quality is the most discussed aspect at 10 positive and 8 negative mentions, while reliability drew 3 negative mentions and no positive ones. The trend is flat to slightly up: 50% approval in September on 36 opinions, then 54% in October on only 12 opinions.

Every number comes from public Hacker News comments collected through the HN Search API. A language model tagged each comment for sentiment, first-hand use, topic and switches; a human spot-checks samples. No vendor can pay for placement.

20,064 comments screened48 about GPT-6 Astra40 first-hand

Read the full methodology →