Meta Llama 4 vs Ternary Bonsai

Meta released Llama 4 Scout and Llama 4 Maverick on April 5, 2026 — the first open-weight natively multimodal models built with a Mixture-of-Experts (MoE) architecture. Scout is a 17B active parameter model with 16 experts that fits on a single NVIDIA H100, with an industry-leading 10 million token context window. Maverick is also 17B active parameters but with 128 experts, delivering performance that benchmarks comparably to GPT-4o and DeepSeek v3 on reasoning and coding tasks. Both models process text, images, and video inputs, and are freely available for download on Hugging Face and llama.com. Llama 4 Scout was trained on 40 trillion tokens of data. The MoE architecture means the models punch well above their weight in active parameter count — Scout competes with models 5-10x its size on many benchmarks, while keeping inference costs low. This release closes the gap between open and proprietary models significantly. Organizations that previously needed to pay for GPT-4o or Claude for multimodal tasks can now run comparable capability locally or via any cloud provider. For the open-source AI ecosystem, Llama 4 is the biggest release of 2026 so far.

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.