Frontend · GitHub · Android

The Open-Source Replacement Plan: $2 for a Beyond Compare Clone, $998 Left to Crowdfund the Next Target

By 朱涛的自习室 · Oct 23, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

A free alternative that ships installers for all three desktop OSes and runs its own 15-minute bug patrol makes a paid diff tool a harder sell.

Summary

A $1000, seven-day experiment has produced the first free open-source stand-in for Beyond Compare. OpenCompare covers folder comparison, code diff, and three-way merge, ships for Mac, Windows, and Linux, and cost roughly $2 in AI usage, leaving $998 of the budget untouched.

The token bill for the week was nearly 900 million. Costs stayed close to zero because the workload ran through a three-tier model stack: Gemini 4 makes top-level decisions, DeepSeek 4.1 Flash handles the bulk of code writing and bug fixing, and the free local Qwen3.8-Flash-Next runs high-frequency checks. A 93%+ cache hit rate means most of that token volume was already paid for at subscription level rather than fresh inference.

After users reported crashes on large folders, the AI converted the comments into issue tickets and pushed three versions (v0.2.2 to v0.2.5) overnight. On a 47 GB directory with 752,000 files, a second comparison now takes 9.3 seconds and memory usage is down 35.8%. OpenCompare now runs automatic maintenance, with an AI patrol on issues every 15 minutes, and the next paid tool to clone is selected by upvotes in the open-rebuilds repo.

Takeaways

— OpenCompare is a free open-source replacement for Beyond Compare, covering folder comparison, code diff, and three-way merge.

— It cost about $2 in AI usage out of a $1000 total budget.

— The effort consumed nearly 900 million tokens over seven days, converted against an OpenCode subscription.

— The model stack is split: Gemini 4 for top-level decisions, DeepSeek 4.1 Flash for roughly 90% of code writing and bug fixing, and local free Qwen3.8-Flash-Next for frequent checks.

— A cache hit rate above 93% keeps the billed cost low.

— User crash reports on large folders were automatically converted into issue tickets and fixed across versions v0.2.2 to v0.2.5 in one night.

— A 47 GB folder with 752,000 files now diffs in 9.3 seconds on a second run, with memory usage down 35.8%.

— OpenCompare ships for Mac, Windows, and Linux, and an AI patrols GitHub Issues every 15 minutes.

— The next software to be rebuilt is selected by upvotes in the open-rebuilds GitHub repo.

Conclusions

The $2 total works only because an OpenCode subscription already absorbed most of the token cost; 900 million tokens at per-token prices would be far more. The trick is tiered assignment and caching, not model price alone.

A 93%+ cache hit rate on a 900-million-token week means the workload is heavily repetitive. Fast, multi-version bug-fix loops are exactly the pattern that makes prompt caching pay off.

Turning user bug reports into tickets automatically and shipping patches overnight removes the maintainer bottleneck that usually decides whether a small open-source tool survives.

Letting community upvotes choose the next paid tool to rebuild is also a demand signal: the targets with the most votes are the ones users actually resent paying for.

Concepts & terms

Three-tier model split

A cost-management pattern that assigns different models by task: a frontier model for rare strategic decisions, a cheap mid-tier model for most code writing, and a local free model for frequent checks. The OpenCompare build uses this split to hold a 900-million-token week near $2.

Prompt cache hit rate

The share of LLM requests served from a stored cache of repeated token prefixes rather than recomputed. A hit rate above 93% means the billed 900 million tokens include a lot of repeated work that costs almost nothing.

Three-way merge

A merge operation that takes a common base version and two divergent versions and reconciles them into one result; standard in version control. OpenCompare offers this as one of its core Beyond Compare replacement features.

OpenCode subscription

An AI coding subscription that bundles token allowances; the 900-million-token week was priced against this plan's conversion rate, which is why the reported spend is about $2.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗