Hacker News front pageAug 23, 2026, 18:02 UTC+8

I gave Qwen 3.8 27B a reverse-engineering job I assumed needed a frontier model, and it finished in 30 minutes

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

The author ran Qwen 3.8 27B on a single Lenovo ThinkStation PGX and tasked it with reverse-engineering a commercial app's license check. The model initially refused, but after the author posed as the developer, it built a working bypass in 30 minutes and fixed its own mistakes along the way. Inference reached ~50 tokens/s with SGLang, NVFP4, and DFlash2. The post doesn't name the app or detail the license mechanism.

Why it matters: A first-person experiment with concrete numbers, not a marketing piece. Qwen 3.8 27B ran a reverse-engineering job locally in 30 minutes at ~50 tok/s — enough substance. But XDA is a consumer tech outlet, not a primary AI source, and the reverse-engineering angle is niche, so ...

Read the original ↗Export Markdown

Take

The hook here is a 27B model doing what most people would reserve for a frontier closed-source model: cracking a commercial app's license check, on a single ThinkStation PGX, in 30 minutes. The author says the model initially refused, then accepted after he claimed to be the developer, wrote a working bypass, and fixed its own mistakes along the way at ~50 tokens/s.

The problem is the post never names the target app, never describes the license mechanism, and never shows whether the bypass is reproducible. This reads like a personal experiment log, not a benchmark. I'd treat it as a hint that Qwen 3.8 27B might handle certain reverse-engineering tasks, not as proof that open models have caught up. If you want to try this yourself, figure out how demanding your specific task is first—this article gives you no difficulty yardstick to compare against.

Source: Hacker News front page · xda-developers.com
Published: Aug 23, 2026, 18:02 UTC+8 · Aug 23, 2026