A benchmark you can inspect and run
Keep the environment, tasks, scoring rules and results in one place. Typed Decision Bench shows the kind of finished output to aim for.
View sample report ↗View source ↗CREATE A WORLD
Describe the work you want a model to do. We design and build its environment, tasks and scoring, ready for benchmarking and an open-model RL gym.
Keep the environment, tasks, scoring rules and results in one place. Typed Decision Bench shows the kind of finished output to aim for.
View sample report ↗View source ↗Choose a model and a training budget. Compare its saved checkpoint with the original on your world and other runnable benchmark samples.
Open RL gym →Pick any of micro1’s expert opportunities and get a working gym: a realistic workspace, tasks from the role’s scope of work and executable checks — run on mock MCP servers or the real services.
Browse job gyms →Read the glassware, fill and tilt from a clip, get a pour spec a GPU simulator runs, and replay recorded runs in 3D against the video’s numbers.
Open Video → World →Structured creation brief
Define your world and how its services should behave.Describe the work, its constraints, and how success should be checked.Studio researches the domain, simulates the systems, mines up to 20 materially distinct graded tasks (only oracle-proven tasks are admitted), and lets you steer the build before release.