Comparisons / Claude Sonnet 4.5 vs Claude Sonnet 4.6
AI models for coding · HN opinions Nov 2025 – Nov 2026
Claude Sonnet 4.5 vs Claude Sonnet 4.6: what developers actually say
Based on 162 opinions on Hacker News from developers who discussed these two, most of them first-hand users.
Claude Sonnet 4.6 comes out slightly ahead on approval, 65% from 65 opinions against 57% from 97, and on code quality, +7 net against +1. Cost mentions are near neutral for both, 4.5's reliability skews negative at +5/-10, and direct preference is tied at 2-2. Sonnet 4.5 has more firsthand reports, 93 against 59. Sonnet 4.5 is the better pick for spec-driven development or long maintenance, and 4.6 for long agentic runs.
When developers explicitly said they prefer one or switched: 2 picked Claude Sonnet 4.5, 2 picked Claude Sonnet 4.6, a tie.
Choose Claude Sonnet 4.5 if
- Want the larger evidence base (93 firsthand reports)
- Value spec-driven development and long maintenance
- Want a model some users switch back to from 4.6
Choose Claude Sonnet 4.6 if
- Want higher approval (65% vs 57%)
- Need speed (+7/-1)
- Run agentic tasks that must last without intervention
Where each one wins and loses
PraiseComplaintsA = Claude Sonnet 4.5B = Claude Sonnet 4.6
Code quality
Reliability
Autonomy
Cost & pricing
Speed
Usage limits
Why developers switched
Picked Claude Sonnet 4.5
Switches back to 4.5 regularly, sees 4.6 as a downgrade, and dreads 4.5 being retired.HN comment · Apr 6, 2026
Finds 4.6 tuned for the shortest answer, which makes exploring and ideating worse than 4.5.HN comment · Apr 6, 2026
Picked Claude Sonnet 4.6
Says 4.6 follows instructions better and writes code closer to their existing production code.HN comment · Mar 12, 2026
Found 4.6 ran three to four times longer than 4.5 without intervention and produced apps on par with Opus.HN comment · Feb 18, 2026
Summaries are paraphrased. Each links to the original comment.
Side by side
| Claude Sonnet 4.5 | Claude Sonnet 4.6 | |
|---|---|---|
| Approval | 57% | 65% |
| Last 90 days | – | – |
| Opinions | 97 | 65 |
| First-hand users | 93 | 59 |