GPT-5.3 Codex: what developers actually say

We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on GPT-5.3 Codex. 51 of the 54 opinions come from people who say they use it themselves.

74%approval, all opinions

–approval, last 90 days

#18of 45 models ranked

51different developers

GPT-5.3 Codex has 74% approval across 54 opinions from 51 authors. Developers praise its code quality (16 positive, 7 negative) and cost (9 positive, 1 negative), and several prefer it to Claude Opus 4.6 or GPT-5.5. Reliability is the weak spot at 3 positive and 4 negative, with reports of long loops and weak tests. Approval was 57% in March 2026, then 72% to 100% in the April to September entries, which have only 2 to 9 opinions each.

Pick it if

Work on backend and systems tasks
Need a low-cost daily driver for scripting and greenfield work
Care most about price for everyday coding

Look elsewhere if

Work mainly in R
Need dependable long autonomous runs
Need strict test coverage and clean refactors

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.

PraiseComplaints

Code quality

Cost & pricing

Reliability

Speed

How opinion has moved

Monthly approval (line) and number of opinions (bars). Months with fewer than 3 opinions are left out.

What developers say

Praise

Considers GPT-5.3 Codex the peak of coding models and misses it for writing good code without overstepping.HN comment · Sep 2, 2026

Says GPT-5.3-Codex was excellent for backend and systems tasks.HN comment · Jul 10, 2026

Says GPT-5.3 Codex balances quality and cost well, cheap enough to use on almost everything.HN comment · Jul 5, 2026

Calls GPT-5.3 Codex possibly OpenAI's best coding model, about 90% of GPT-5.5 at a fraction of the cost.HN comment · Jun 30, 2026

Used Codex-5.3 as a daily driver and got what they wanted quickly for scripting and greenfield projects.HN comment · Jun 20, 2026

Says Codex-5.3 was far better than GPT-5.5 for them, but OpenAI removed it from their subscription.HN comment · Jun 20, 2026

Finds GPT-5.3-Codex does a better job than Opus 4.6/4.7 at a fraction of the cost.HN comment · May 29, 2026

Uses GPT-5.3 Codex for straightforward coding; capable enough and much cheaper than GPT-5.5.HN comment · May 18, 2026

Complaints

Had poor results with GPT-5.3 Codex, so used GPT-5.2 for design and 5.3 Codex only for execution.HN comment · Jun 20, 2026

Says GPT-5.3-Codex falls short on R programming work.HN comment · Apr 15, 2026

Started quickly but got stuck in a five-hour loop that made no further progress on the clone task.HN comment · Apr 2, 2026

Describes Codex 5.3 injecting a CSS hack instead of properly refactoring the markup.HN comment · Mar 12, 2026

Says Codex 5.3 wrote hollow tests and overstated test coverage during refactors.HN comment · Mar 10, 2026

Says GPT-5.3 Codex on xhigh makes mistakes in 25-50% of commits and plans tasks poorly.HN comment · Mar 6, 2026

Says GPT-5.3 Codex uses jargon and doesn't clearly explain its plans, unlike Opus.HN comment · Mar 6, 2026

Says Codex 5.3 asks to continue every few lines, burning far more credits than Opus in Copilot.HN comment · Mar 5, 2026

Summaries are paraphrased. Each links to the original comment.

GPT-5.3 Codex compared

GPT-5.3 Codex vs Claude Opus 4.6 GPT-5.3 Codex vs GPT-5.4

FAQ

Is GPT-5.3 Codex worth it?

Most opinions are positive: approval is 74% across 54 opinions. Code quality and cost score well. Reliability is mixed at 3 positive and 4 negative comments.

Is GPT-5.3 Codex better than Claude Opus 4.6?

Five developers preferred GPT-5.3 Codex to Opus 4.6, and two preferred Opus 4.6. Cost is a common reason to pick Codex, though some said quality falls short on harder work.

Is GPT-5.3 Codex reliable?

Reliability comments lean negative at 3 positive and 4 negative. Reports describe a five-hour loop with no progress and tests that overstated coverage.