ASI Research

ASI Research is an AI research lab working toward Autonomous Superintelligence. Our route there is automating research, above all AutoResearch, automating AI research itself, so that improvements compound instead of being hand-built one at a time.

That goal sets the agenda. It needs model architectures that support long context and enable continual learning: attention mechanisms and what they cost at length, positional encodings and how structure is represented, fast-weight and delta-rule updates that let a model keep learning after pretraining ends. An agent that cannot hold a long rollout in context, or cannot carry what it learned in one episode into the next, is limited by its architecture, and no amount of environment design fixes that.

And it needs agentic reinforcement learning at scale, so our models can carry out long-horizon tasks. Capability now comes from what an agent is trained against, not only from the model it starts as, so the training signal, the environments, and the infrastructure behind both are the research problem. We build RL environments at scale, terminal (TUI) and graphical (GUI), with executable ground truth rather than surface matching, because an environment whose reward can be gamed teaches the model to game it.

We also build our own harness, the scaffolding through which the model perceives, acts, and is evaluated. We make the model and harness coevolve: a stronger model demands a better harness, and a better harness exposes what the model should learn next. This loop is how we get to recursive self-improvement.

We keep the group small and the agenda long. We publish what we learn and open-source what we build. If our questions are your questions, we would like to hear from you.