Arcee Trinity-Large-Thinking vs Ternary Bonsai

Arcee AI, a 30-person startup, has released Trinity-Large-Thinking — a 399B sparse mixture-of-experts reasoning model under Apache 2.0. Only 13B parameters activate per token, giving it inference speed 2-3x faster than comparable dense models. In internal benchmarks and early community testing, it ranks #2 on PinchBench, trailing only Anthropic's Opus 4.6, at a list price of $0.90/M output tokens — roughly 96% cheaper than frontier closed models. The model was trained in a $20M, 33-day run on 2,048 NVIDIA Blackwell GPUs. Arcee trained it using a constitutional AI-style process with synthetic chain-of-thought data generated from multiple frontier models, then applied a reinforcement learning phase using outcome-based rewards on math, code, and logic benchmarks. Trinity-Large-Thinking is the strongest open-weight reasoning model released to date on a commercial-friendly license. For companies with privacy requirements or custom deployment needs, it represents a credible alternative to frontier closed APIs — especially for code generation, mathematical reasoning, and structured data tasks where the gap between open and closed models has historically been widest.

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.