Qwen3 Family vs Ternary Bonsai

Alibaba's Qwen team released the full Qwen3 model family this week — 8 models ranging from 0.6B to 235B parameters, spanning both dense and Mixture-of-Experts (MoE) architectures. The headline model is Qwen3-235B-A22B, a 235B MoE that activates 22B parameters per token and matches GPT-4.1 on coding and math benchmarks while running at a fraction of the cost. All Qwen3 models feature switchable "thinking modes" — a built-in chain-of-thought toggle that can be enabled or disabled per request. This eliminates the need for separate reasoning vs. instruct variants, letting developers trade latency for accuracy dynamically. All models are released under Apache 2.0, with weights available on Hugging Face and ModelScope. The smaller models are competitive at their size class: Qwen3-4B reportedly matches Qwen2.5-72B-Instruct on several benchmarks, and the 0.6B model is designed to run efficiently on embedded and edge devices. The release also introduces a new multilingual benchmark covering 119 languages, on which the Qwen3 family sets new state-of-the-art scores for open-weights models.

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.