GPT-5.6 Sol: what developers actually say

We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on GPT-5.6 Sol. 210 of the 247 opinions come from people who say they use it themselves.

67%approval, all opinions

66%approval, last 90 days

#19of 45 models ranked

206different developers

GPT-5.6 Sol has 67% approval across 247 opinions from 206 authors, and 210 of those opinions are firsthand. Code quality is the clearest strength at 73 positive mentions against 26 negative, and speed is 14 against 2. Reliability is weaker at 10 positive against 16 negative, and users report context limits, weekly quotas drained in a day, and conceptual mistakes. Approval was 70% in July, 72% in August, 56% in September, then 73% in October on 22 opinions; the 90-day figure is 66%.

Pick it if

Want strong code quality, with 73 positive mentions against 26 negative
Want faster output, with 14 positive speed mentions against 2 negative
Do web development, deep research, or code review

Look elsewhere if

Need consistent reliability, where mentions run 10 positive to 16 negative
Work in large codebases, where mentions run 3 positive to 5 negative
Need generous usage limits, with reports of weekly quotas gone in a day

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

Autonomy

Speed

Usage limits

Large codebases

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

Says Opus once felt much worse to use than Sol, before Anthropic's fix.HN comment · Oct 28, 2026

Says the whole team switched to Codex with Sol instantly when it arrived.HN comment · Oct 28, 2026

Says they moved from Claude to GPT-5.6 Sol and preferred its prose.HN comment · Oct 26, 2026

Says they mostly use GPT-5.6 Sol and will test Terra against cheaper models before drawing conclusions.HN comment · Oct 26, 2026

Says they generally like Sol.HN comment · Oct 23, 2026

Says Sol is better for web dev, deep research, and code review.HN comment · Oct 20, 2026

Says they liked Sol's coding results, though the new DeepSeek release should narrow that advantage.HN comment · Oct 17, 2026

Says GPT-5.6 Sol finished a code review in about 5.5 minutes with two relevant high-priority findings.HN comment · Oct 17, 2026

Complaints

Says Sol deviates from prompts by taking independent actions, causing data loss while cleaning up a git repository.HN comment · Oct 4, 2026

Says GPT-5.6 Sol's context window is too small and hit limits too fast, so cancelled.HN comment · Sep 30, 2026

Says Sol was acceptable but often produced buggy UI code.HN comment · Sep 29, 2026

Says Sol used up their entire weekly allowance in a single day.HN comment · Sep 29, 2026

Says Sol drained the $100 plan's weekly limits in a single day.HN comment · Sep 29, 2026

Says GPT-5.6 Sol still makes conceptual mistakes and misses edge cases, needing plans corrected.HN comment · Sep 26, 2026

Says Sol has made increasingly more mistakes over the past two weeks, even in reviewer roles.HN comment · Sep 22, 2026

Says Sol's intelligence has declined markedly, even when run at high effort settings.HN comment · Sep 22, 2026

Summaries are paraphrased. Each links to the original comment.

GPT-5.6 Sol compared

GPT-5.6 Sol vs Claude Fable 5 GPT-5.6 Sol vs Claude Opus 5 GPT-5.6 Sol vs GPT-6 Astra

FAQ

Is GPT-5.6 Sol worth it?

It has 67% approval from 247 opinions. Code quality and speed score well, while reliability and usage limits draw more complaints. It fits developers who value output quality and can manage quota use.

Is GPT-5.6 Sol better than Claude Fable 5?

In head-to-head mentions, 10 people preferred Sol and 4 preferred Claude Fable 5. Sol also led Claude Opus 5 by 7 to 3. Against GPT-6 Astra, 3 preferred Sol and 4 preferred Astra.

How reliable is GPT-5.6 Sol for coding?

Reliability mentions are 10 positive and 16 negative. Complaints cite conceptual mistakes, missed edge cases, a small context window, and usage limits reached quickly. Some users also report buggy UI code and a decline in quality over two weeks.