60.63M parameters
Compact enough to study specialization without relying on massive model scale.
BitVision research program
Learning systems built around measurable decisions.
Explore BitVision self-play research and run the BitVision-Cell policy directly in your browser. Public demonstrations document results and limitations separately from architectural goals.
Open BitVision-Cell dashboardBitVision learning model
Highly focused learned behavior at 60.63M parameters.
BitVision is Global Intent Company's proprietary learning-model research program focused on a simple question: how capable can a compact model become when its training, feedback, and environment are aligned around a clear intent?
In self-play, BitVision learns to make decisions, pursue objectives, adapt to an opponent, and improve measurable task performance without needing to become a general-purpose model. The architecture and implementation details remain proprietary.
New live experiment
BitVision-Cell v0.2 runs an 8.75M-parameter checkpoint directly in the browser and lets the learned policy control a synthetic tissue/pathogen environment in real time.
Launch live checkpoint inferenceIntent-focused
Optimized around defined behavior and measurable objectives instead of broad general knowledge.
Learning through self-play
Behavior emerges through repeated interaction, competition, feedback, and evaluation.
Proprietary research
Public results show what the model can do without exposing the architecture that makes it work.
Learning in motion
Recorded BitVision self-play development run
Model size
60.63M
Training mode
Self-play
Focus
Specialized intent
Demonstration footage is presented to show observed behavior and training progress. Internal model architecture, representation, and optimization details are intentionally not disclosed.