Crush: what developers actually say
We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on Crush. 19 of the 26 opinions come from people who say they use it themselves.
Developers rate Crush positively overall: 77% approval across 26 opinions from 18 authors, 19 of them firsthand. UX drew 5 positive and 2 negative mentions, and customization drew 3 positive and no negative ones. The trend leans positive, from 50% approval in late 2025 to 100% in each month from June through September 2026, though those months hold only 1 to 3 opinions each. People praise the single Go binary, easy model switching, and low configuration, while complaints center on missing plan mode and subagents, weak copy-paste in the terminal, and a proprietary fork of OpenCode.
Pick it if
- Want one Go binary that is easy to reason about
- Switch models mid-task often
- Want a terminal agent that rarely needs configuration
Look elsewhere if
- Need plan mode and subagents today
- Copy and paste from a terminal constantly
- Want a fully open-source codebase
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.
Crush compared
Best Crush alternatives, according to developers who switched →
FAQ
Is Crush worth it?
Most developers in the data say yes. Approval is 77% across 26 opinions, and 19 come from people who used it directly. One user reported about $8 in spend over seven weeks. Recent months show 100% approval, but on 1 to 3 opinions each.
Is Crush better than OpenCode?
Two developers preferred Crush over OpenCode, and one preferred OpenCode instead. One called Crush a single Go binary that is easier to reason about than OpenCode. Another said Crush is a fork of OpenCode and criticized it as proprietary.
Does Crush work with local models?
Developers report it works with a local llama-swap server over Tailscale and with any provider. Several said the docs for local LLM and arbitrary API setup are scattered or thin. One said it works reasonably well only for simple tasks.