Refresh
Simulated worlds with verifiable rewards to train coding and computer-use agents.
Refresh provides simulation environments and evaluation tasks for coding and computer use, designed for training and benchmarking AI agents. Its environments are instrumented so task success is measured automatically through verifiable rewards, supporting both supervised fine-tuning and reinforcement learning. Verifiers range from deterministic checks, such as passing test suites or correctly filled form fields, to rubric-based scoring for less binary tasks.
The product spans three environment types: high-fidelity clones of connected software suites, 3D animation and simulation work inside Blender, and coding tasks targeting economically valuable professional work. Its computer-1 harness runs the same task across multiple models for comparison, and the company contributes to Harbor and reviews tasks for Terminal-Bench. Enterprises can convert their own workflows into measurable, repeatable tasks and iterate against an objective signal.
Refresh works with frontier AI labs and enterprises. Environments and bespoke datasets are requested directly by contacting the company; no free plan, trial, or pricing structure is stated.
12 alternatives to Refresh
Ranked by how well each tool replaces Refresh: shared features, audience, price and popularity.
Provides live environments for training AI agents, supplying real-world MCP environments,
Covers 1 of 15 key features.
55 out of 100 match—Platform for building UIs, automating workflows, and deploying AI agents.
Covers 0 of 15 key features, has a free plan and is open source.
Free planOpen source54 out of 100 match$29/moHelping the world's largest organizations safely refactor their operations to unlock new-g
Covers 0 of 15 key features.
54 out of 100 match—The Platform for Production-Ready Agentic Workflows
Covers 0 of 15 key features, has a free plan and is open source.
Free planOpen source54 out of 100 match$49.17/moBuild powerful AI agents for yourself, your team, or your enterprise — no coding required.
Covers 0 of 15 key features and has a free plan.
Free plan53 out of 100 match$20/moTesting sandboxes, built for AI agents
Covers 0 of 15 key features and has a free plan.
Free plan53 out of 100 matchUsage-basedAI Systems Built for the Enterprise
Covers 0 of 15 key features and has a free plan.
Free plan52 out of 100 matchUsage-basedDatasets and environments to help healthcare agents succeed at complex multi-step tasks
Covers 3 of 15 key features.
52 out of 100 match—