Teortaxes Notes DeepSeek Fluency in Russian

Pseudonymous AI developer Teortaxes posted that DeepSeek remains the only Chinese model family producing fluent Russian. The account stated that models such as K3 and 5.3 shift into awkward Runglish while Flash avoids this pattern. Teortaxes linked the difference to pretraining data quality and added that several other families including MiMo, Hy, Doubao and Qwen had performed poorly on the same task in earlier checks. The post frames the observation as a repeated finding rather than a new benchmark result. No independent verification or additional model tests appear in the supplied evidence.