GigaChat Audio AI, an updated large language model from Sberbank, can now process audio files and voice messages without converting speech to text first. The model detects users’ positive or negative emotions based on intonation, voice characteristics, and pronunciation nuances, allowing it to respond more gently to an irritated user or mirror the mood of someone sharing good news. It can process recordings up to three hours in length, navigate within them, distinguish between different speakers, and provide summaries with timestamps. The AI assistant can also remember facts from voice dialogues and rely on them in future sessions, with users able to view, edit, or disable this memory function in profile settings. According to in-house tests, GigaChat Audio secured a win rate of 70% on the Arena-Hard-Audio benchmark, almost on par with Gemini 3 Flash preview (77.5%) and higher than Gemini 2.5 Pro (62%). In terms of emotion recognition accuracy, Sber’s model is at 80%, outperforming Qwen3-Omni-30B (70%) or Kimi-Audio (62%).