Horizon
Horizon supplies RL training gyms and evaluations for reasoning, tool use and computer use, using WorldSim to simulate enterprise environments such as GitLab, Jira, CRM, email and chat and to produce calibrated reward and preference signals for post-training.
What it does
- Model training
- Evaluation and observability
- Data labelling
- Agent orchestration
Features 2
- WorldSimRecreates enterprise environments such as GitLab, Jira, CRM, email and chat with realistic business data.
- Preference labelsTurn human judgment into structured comparisons across agent trajectories.
Sources
- Pricing
-
Official documentation · 5 Sep 2026
Fetched. No prices, tiers or rates on the page.
- Description
-
Official documentation · 5 Sep 2026
Fetched. H1 'Environments for post-training, at scale'; page calls it 'RL training gyms & evals for reasoning, tool use, and computer use' and names WorldSim and the enterprise apps it simulates.
- Sold within
-
Live re-check 2026-09-15: the page never mentions Recursion, Terra or Alignerr as peers (contradicting the prior note) and reads as a feature within Labelbox's software rather than a clearly independent peer product, with no dedicated pricing or CTA for Horizon itself.
- Interfaces
-
Platform listing · 14 Oct 2026
npm listing published by Labelbox (npm scope @labelbox, maintainer labelbox-operator): 'Command-line client for Horizon'.
- Intended for
-
Named customers Meta, Google Cloud, Intuitive Surgical; targets 'GUI-based task execution across enterprise software' and enterprise knowledge work domains.
Also from Labelbox
-
Recursion Managed Agents Agent platform
Managed agents for engineering and research teams: runs fleets of agents on a team's code, infrastructure, experiments and evals, with every run graded and remembered.
-
Terra Data service
Terra is Labelbox's robotics data product delivering video, trajectories and multimodal annotations across pre-training, post-training and evaluation stages, including expert teleoperation with action labels and multiple camera perspectives for embodied foundation models.
-
Alignerr Data service
Alignerr is Labelbox's expert-network product that routes AI training and evaluation tasks to credentialed contributors across 200+ knowledge domains and 40+ countries and returns structured outputs for RL training, RLHF and evaluation workflows.
-
Catalog Application
Catalog is Labelbox's data curation and search product providing out-of-the-box search across images, text, video, conversations and documents over metadata, vector embeddings and annotations without building your own vector database infrastructure.
-
Annotate Application
Annotate is the data labeling product within Labelbox, providing 10+ built-in editors for multimodal chat, LLM evaluation, prompt/response generation, computer vision and NLP, plus customizable labeling and review workflows and team performance monitoring.
-
Labelbox Foundry Developer tool
Foundry runs third-party foundation models over data already in Labelbox to pre-label and enrich image, text and document datasets without code, routing the predictions to human review; billed as inference cost per model run plus Labelbox Units.
Open Labelbox in the directory
Something wrong here? Send a correction — quote this product id: labelbox-horizon.