Scaling intelligence density
Syzygy is a research lab dedicated to building more capable models that need less memory, compute, and energy. We believe that the future of intelligence is efficient, locally controlled, and owned rather than rented.
Introducing
Mach-2 Additive Medium
Qwen3.8-Flash-Next at 1.70 bits per weight, inferencing solely with addition rather than multiplication. 125.7 billion text parameters in 26.7 GB, retaining 96% of the BF16 model's benchmark scores and fitting on a single 32 GB GPU.
- 1.70 bits
- per weight
- 125.7B
- text parameters
- 96%
- average retention vs BF16
- 26.7GB
- on disk
Our mission
Today, frontier intelligence is limited to datacenters. In addition to skyrocketing economic costs, this paradigm raises structural concerns over privacy, autonomy, and sustainability (economic, societal, and environmental). What does the world begin to look like when the power of superhuman intelligence is so tightly concentrated?
We believe that societally beneficial, cost-efficient, and outcomes-driven AI becomes structurally feasible when users are able to own intelligence as easily as an operating system. Once intelligence is owned rather than rented, we predict incentives will shift from metering units of reasoning to completing as much useful work as possible.
To that end, our mission is to define a new frontier of efficiency by building models that provide more intelligence for less energy.
Syzygy
/ˈsizəjē/ · noun
Our mission doesn't end with models
Efficient intelligence requires co-design of hardware, runtime, and harness. Squeezing every last drop of intelligence from the model doesn't matter if the stack around it isn't built with the model in mind.
Join us
We are a team of mathematics and physics researchers building AI as we think it should be. If our mission resonates, please reach out.