RL training gyms and evals for reasoning, tool use and computer use
Horizon is Labelbox's reinforcement-learning environment product for post-training frontier models. It generates RL environments and evals with calibrated reward signals targeted at pass@k goals across knowledge-work domains, including WorldSim, which recreates enterprise environments such as GitLab, Jira, CRM, email and chat with realistic business data so agents can practise across hundreds of MCP tools on computer and terminal tasks.
For people
For agents
Nothing listed yet.