Qwen3.5-Omni
#52 v Multimodální modelyalibaba · v3.5 · omni · od 2026-03-30 · 20× · naposledy 30. 6. 2026
7
Momentum
Qwen3.5-Omni is a native omnimodal large language model from Alibaba's Qwen team, combining hundreds of billions of parameters in a Hybrid-Attention Mixture-of-Experts (MoE) architecture. The model natively processes text, images, audio, and video in a single inference pipeline and generates both text and real-time speech output. It is distributed exclusively as a proprietary cloud API service via Alibaba Cloud Model Studio (DashScope) — no model weights are publicly available. The model comes in three variants: Plus, Flash, and Light.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Context Window (Tokens) | 256,000 tokens (all three variants: Plus, Flash, Light) – equivalent to >10 hours of audio or >400 seconds of 720p video at 1 FPS |
| License | Proprietary (closed-source); no public model weights – access exclusively via Alibaba Cloud API (DashScope / Model Studio) and Qwen Chat |
| Platform | Alibaba Cloud Model Studio (DashScope) – Offline API & Realtime API; OpenAI-compatible endpoint; also accessible via Qwen Chat (chat.qwen.ai) and Hugging Face demo |
| Release Date | March 30, 2026 |