Pioneering the next leap in local AI intelligence.
Creating stunning AI models that are highly capable, yet easy to run, with a focus on reasoning capabilities as well as coding.
The Parallax models are powerful on their own, but Parallax Squadron takes them to the next level. This framework allows a commander model to deploy parallel sub-agents of itself, delegate workloads, and orchestrate complex tasks across your local hardware—all while drastically improving overall efficiency.
Our first real step. We proved that a small, highly optimized model could mimic Opus-style reasoning. This 4B model validated our distillation techniques and showed the world that consumer GPUs could handle complex coding and agentic workflows.
Extreme engineering. We pushed the boundaries of tiny to see how small we could go before reasoning broke. The answer was 800 million parameters. It runs on almost any hardware, easily and fast, bringing local AI to the masses.
The current flagship. We stopped messing around. With 9 billion parameters and a 262K context window, it proved that local hardware could handle the heavy lifting. Trained on 150K+ elite traces, it competes with the giants while running entirely offline.
The finalized revision of the Parallax 2 family's mini variant. Bringing Local AI to the masses with an ultra-small model that can run on anything. Parallax 2 Mini is more powerful than Parallax 2 Preview, while maintaining the same 4 billion parameter size you know and love!
Our next leap. A super-flagship designed to completely reduce the need for an AI subscription to nothing. We are scaling parameters and refining reasoning to push local inference to its absolute limits. Big AI companies will be left behind.