AI article
Best LLM for Translation? I Tested Six Models on 46 Manga Pages
Community description: Six LLMs, the same 46 manga pages, 278 lines, every mistake counted. Two models made none. A light model beat a middle one, and most mistakes were not about vocabulary.
Dev.to | Oct 10, 2026 | Peter
Automated excerpt
One run on 46 pages shows the mistakes I saw. Model size is not a ranking. claude-sonnet-5-5, the middle model of its family, made five mistakes. gemini-3. 8-flash, a light model, made none. One run per model, 46 pages, one judge (me), three source languages, English only as the target.
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.