Shizuku-Duplex Adds Full-Duplex Japanese Conversation
Japan-based Shizuku AI announced Shizuku-Duplex, a full-duplex model designed for Japanese conversation that can begin processing before a request is complete. The model combines a foreground interaction model with a background model for asynchronous complex requests, supports audio, text-monologue and action outputs, and uses 80-ms interaction frames with context compression every five minutes. Shizuku AI says it achieved the highest conversational naturality among six tested Japanese models and ranked second to OpenAI’s GPT-Live in response quality, but no test method, scores or independent verification were provided.
Shizuku-Duplex introduces a new full-duplex design for Japanese conversational AI and is also serving as the basis for Shizuku AI’s planned companion AI, 雫.