Announcing Huzzle World · #1 on ScreenSpot-Pro at 2B and under at ≤2B Read the results →
RL Environments Enterprise Huzzle World Talk to founders opens Calendly

Where AI makes contact with the real world.

We build RL environments for frontier AI labs, and the evaluations and custom models that put AI to work inside the enterprise.

Huzzle Labs is behind Huzzle World, our family of computer-use models at 1B to 8B, built on a proprietary architecture, our own RL environments and training from video. HuzzleWorld-2B is #1 on ScreenSpot-Pro among all models of 2B and smaller, and single-pass it outscores grounding models up to 36 times its size.

#7
Fastest-growing company in the UK
Sifted · 2026
#14
Fastest-growing company in Europe
Nebius & J.P. Morgan · 2026

Huzzle Labs is one of the leading AI platforms providing post-training data and infra for AI labs.

Emre Guven
Meta Superintelligence Labs
Work with us

Training frontier models, or putting them to work in production? Let’s talk.

→ HuzzleWorld Computer-use models → RL Environments For AI labs → Enterprise Custom evals & models Talk to founders

Benchmarks

We are building benchmarks that test real professional work, starting with InsureBench, a proposed benchmark that measures how models perform on claims management and underwriting workflows.

Enterprise insurance reasoning
InsureBench In progress

A benchmark for how language models handle real insurance work. It spans two core workflows — underwriting and claims & coverage — built from document-grounded cases that each resolve to a single verifiable answer. Models are evaluated pass@1 and scored against the recorded outcome, not the wording of the response.

Leaderboard Methodology Underwriting Claims

Backed by 10X Founders·Angel Invest·Emerge·a16z Scout Fund·Thomas Wolf Hugging Face·Bernd Heinemann Allianz·Yaser Khalighi Stanford