Blog
Benchmarks, evaluations, and datasets for training frontier AI models. Original research on simulation engines for coding, healthcare, and document automation.

Refresh is an AI infrastructure company that builds reinforcement learning environments for coding.
refresh.devCrawled Oct 16, 2026
Benchmarks, evaluations, and datasets for training frontier AI models. Original research on simulation engines for coding, healthcare, and document automation.
An SFT dataset of real developer workflows for training production-grade code models.
An RLVR dataset curated from Gauntlet, delivered in Harbor format for reinforcement learning.
The company tested three frontier AI models on 70 real-world PDF form-filling tasks.
Refresh is hiring ex-founders, GTM engineers, and research engineers building the simulation engines frontier labs train on.
The company builds environments that frontier labs train on.
The role oversees end-to-end delivery of strategic projects for computer-use and coding data builds, including scoping work with frontier-lab customers and meeting timelines.
The role manages the technical interface between Refresh and the frontier labs it sells to, including model evaluations, pilot environments, sales demos, and product feedback.
The role involves building benchmarks for frontier labs to measure real-world coding and computer-use capability, translating expert workflows into evaluations, and testing them against frontier models.
Refresh builds simulation engines for coding and computer use, for frontier labs to train the next generation of virtual software engineering collaborators.
The privacy policy describes how the company collects, uses, and protects information.
The terms set rules and guidelines for using the website and services.