GPT-5.4: what developers actually say
We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on GPT-5.4. 101 of the 114 opinions come from people who say they use it themselves.
GPT-5.4 holds 72% approval across 114 opinions from 105 authors, and code quality is the clearest strength at 37 positive mentions against 14 negative. Developers praise its task adherence, context compaction and price, with cost drawing 12 positive mentions and 2 negative, and 6 positive mentions on large codebases with no complaints. The trend is uneven: April 2026 carried most of the data at 61 opinions and 75% approval, while June through September total 21 opinions at 91% to 100% approval, so the recent rise rests on thin samples. Complaints focus on forgotten instructions, missed bugs in review, wasted effort checking diffs and weak orchestration of large workflows.
Pick it if
- Want strong code quality on large repositories
- Need steady task adherence through long context compactions
- Care about price at high usage volume
Look elsewhere if
- Need reliable orchestration of large multi-file workflows
- Want security analysis on your own repositories
- Expect accurate bug detection during code review
Strengths and weaknesses
Each row counts opinions that mentioned the topic. Bars to the right of the line are praise, bars to the left are complaints.
How opinion has moved
What developers say
Praise
Complaints
Summaries are paraphrased. Each links to the original comment.
GPT-5.4 compared
FAQ
Is GPT-5.4 worth it?
Most developers in this sample say yes, with 72% approval from 114 opinions. Cost is a commonly cited strength at 12 positive mentions against 2 negative. The main risk is instruction drift and uneven reliability over long sessions.
Is GPT-5.4 better than Claude Opus 4.6?
Sentiment leans toward GPT-5.4, with 10 people preferring it over Claude Opus 4.6 and 4 preferring Opus 4.6. That is opinion data only, with no benchmark results in the sample.
Is GPT-5.4 good for large codebases?
The data supports it, with 6 positive mentions for large codebases and none negative. Some users still report inconsistent orchestration on a 200-file QA workflow, so pair it with deterministic tooling for very large jobs.