Gemma 4 vs Ternary Bonsai

Gemma 4 is Google DeepMind's fourth-generation open model family, released April 2, 2026, under Apache 2.0. Four variants ship in the family: E2B and E4B edge models that run fully offline on phones, Raspberry Pi, and NVIDIA Jetson; a 26B Mixture-of-Experts model that activates only 3.8B parameters at inference; and a 31B Dense flagship. The 31B scores 1452 on the Arena AI text leaderboard (third among all open models), hits 89.2% on AIME 2026 math, and 85.2% on MMLU Pro — versus Gemma 3's 20.8% on AIME. All four model sizes accept text and image inputs. The edge models additionally handle native audio and video, making them the first on-device models with full multimodal coverage. Context windows reach 256K tokens on the large variants, enabling entire codebases or long documents in a single prompt. Native support for tool use, structured output, and agentic workflows is baked in from the start. For the open-source AI community, Gemma 4 is a watershed: a commercially permissive model that genuinely competes with closed-source alternatives on reasoning benchmarks. Gemma downloads crossed 400 million before this launch — Gemma 4's edge deployment story, combining on-device inference with frontier-class reasoning, looks set to make that number look small.

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.