After evaluating several models for translating Chinese subtitles into English, a user has identified Qwen 3.5 27B as the most effective solution.

Performance of Qwen 3.5 27B

According to the tester, Qwen 3.5 27B outperforms other models up to 70B in terms of translation quality. Their local setup, equipped with 24GB of VRAM, produces translations with a tone and consistency comparable to those of GPT-3.5 and Gemini. MoE (Mixture of Experts) models and thinking only models have shown inferior performance in this specific use case.

For those evaluating on-premise deployments, there are trade-offs to consider carefully. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.