Fundamental Research

Building foundation models
from the ground up.

We don't just fine-tune or wrap existing intelligence. At PlusB Intelligence Lab, we engineer native architectures designed for a post-cloud future.

/ Our Philosophy

Intelligence shouldn't require supercomputers.

Accessible Hardware

Designed to run flawlessly on low-quality, consumer-grade hardware. We heavily optimize inference pathways to democratize powerful AI capabilities without heavy infrastructure dependencies.

High Intelligence

Despite the reduced parameter footprint, our models demonstrate emergent reasoning and high comprehension typically reserved for massive data-center models.

Extremely Low Cost

By rethinking the fundamental transformer architecture and memory bounds, we drop the computational cost of intelligence by orders of magnitude for sustainable scaling.

Current Iterations

Status: Active Training
Treevik 1
Base Generation
v1.0

Our first proof-of-concept for highly-efficient intelligence. Proved that careful data curation and novel attention mechanisms could match much larger state-of-the-art systems on core reasoning tasks with a fraction of the hardware.

Focus
General Reasoning
Hardware Profile
Edge & Consumer GPU
Treevik 2
Next Generation
v2.x LATEST

A significant architectural leap. Achieves state-of-the-art conversational dynamics and logical deduction while retaining the low-cost, low-hardware footprint established by its predecessor. Introduces dynamic context routing for improved multi-turn reasoning.

Focus
Complex Logic & Context
Hardware Profile
Optimized Edge

Research Direction

"Our ultimate objective is a ubiquitous intelligence layer—models that are so lightweight they can be embedded everywhere effortlessly, while remaining capable enough to solve complex, real-world problems."