Create model candidates
Foundation Models
Evolving the models themselves — merging fine-tuned expert models into stronger ones without further fine-tuning.
EvoGM · Evolutionary model mergingEnabling intelligence to adapt, create, and evolve across digital and physical worlds.
Durable per-agent checkpoints connect task recovery, visible execution phases and bounded tool concurrency in Genesis 0.14.3.
Choose a project, approve a concrete action and follow background development without tying its lifetime to the chat.
Sharing weights makes architecture search affordable. EvoNAS tackles the harder question: whether the resulting scores can guide us toward a vision model worth deploying.
01Why Evolution
Evolution is nature's most proven way to create intelligence. EvoX brings that loop to AI: generate many candidates, test them for real, keep what works, and build on it — so capability compounds through experiment instead of hand-tuning.
Create diverse candidates; in later rounds, inherit and vary retained solutions.
02Applications
From models to agents, robots, and scientific discovery — one evolutionary loop, applied wherever intelligence needs to adapt.
Create model candidates
Evolving the models themselves — merging fine-tuned expert models into stronger ones without further fine-tuning.
EvoGM · Evolutionary model mergingThe root agent branches from main
Recursive agents that decompose, build, and evolve entire codebases — from a blank repository to shipped software.
Genesis · Recursive agentic systemPolicies control stride, lift, balance and timing
Evolving robot control policies through massively parallel interaction with simulated environments.
EvoRL · Evolutionary reinforcement learningPropose airfoils with different camber and thickness
Data-driven black-box optimization for scientific and engineering design, validated on supercritical airfoils.
EvoGO · Supercritical airfoil design03Hardware
Turn batches of complex tasks into tensor computation.