Qwen3.5-Omni vs Ternary Bonsai

Qwen3.5-Omni is Alibaba's most advanced multimodal model yet — a native Thinker-Talker architecture that processes and generates text, audio, and video in a single unified system. Released in three variants (Plus, Flash, Light), it supports a 256k context window, 10+ hours of audio, and 400 seconds of 720p video at 1 FPS, with speech recognition across 113 languages and dialects. The headline capability is what Alibaba is calling "Audio-Visual Vibe Coding" — an emergent behavior where the model writes functional code based solely on watching a video and listening to spoken instructions. In demos, it takes a hand-drawn sketch held up to a camera and converts it into a working React webpage in real time. This wasn't an explicitly trained capability; it emerged from the model's unified multimodal architecture. The model uses semantic interruption and turn-taking intent recognition for real-time interaction, and TMRoPE for temporal multimodal position encoding. The catch: Alibaba broke from its open-source streak and kept Qwen3.5-Omni proprietary, accessible only through their chatbot interface and Alibaba Cloud. The open-source community has noticed — and is not pleased.

PrismML's Ternary Bonsai is a family of ultra-compressed language models using 1.58-bit weights — meaning every parameter is stored as -1, 0, or +1, with no higher-precision layers anywhere in the architecture. The line-up covers 8B, 4B, and 1.7B parameter models. The flagship 8B model fits in 1.75 GB of RAM, a 9x reduction versus a 16-bit baseline. Unlike earlier 1-bit experiments that felt like a party trick with serious capability regressions, Ternary Bonsai 8B outperforms PrismML's own prior 1-bit Bonsai 8B by 5 points on average across standard benchmarks. The team also ships WebGPU inference, so the 1.7B model runs entirely in a browser tab. This is the first time a production-quality chat model has run with no server at all. The real-world use case is edge and offline deployment: medical devices, air-gapped government systems, consumer apps that need to work without a signal. At 1.75 GB, the 8B model fits on the GPU RAM of a six-year-old gaming laptop. PrismML is positioning this as the foundation for truly offline AI — a credible claim if the capability benchmarks hold up under real-world testing.