Benchmark
Measure what models can do in real professional work.
We are building benchmarks for 3D generation and long-horizon financial work, combining runnable verification scripts with expert-defined task distributions and rubrics.
Featured benchmarks
Two benchmarks in development.
3D
A verifiable measure of 3D scene and asset generation.
The 3D benchmark combines multi-file projects, spatial reasoning and runnable verification scripts to turn “was the 3D result generated correctly?” into automatically checkable criteria.
- Multi-file projects for 3D scenes and assets
- Spatial reasoning about structure and relationships
- Runnable verification scripts with automatically checkable criteria
Finance Long-Horizon Agent
Evaluate agents on long-horizon financial work in Chinese contexts.
Following the Finance Agent Benchmark direction, we focus on Chinese financial conventions, compliance workflows and collaboration across roles. Tasks are open-ended, with office documents as deliverables; finance experts define the task distribution and evaluation rubrics.
- Chinese financial conventions and compliance workflows
- Multi-role collaboration across long-horizon tasks
- Open-ended tasks delivered as office documents
- Task distributions and rubrics defined by finance experts
How we evaluate
One method behind every benchmark.
Real tasks, real deliverables
Every task mirrors work a professional is paid to do. Models are scored on complete deliverables — documents, code, analyses — not multiple-choice answers.
Expert-authored rubrics
Domain experts define the scoring dimensions and per-task anchors, so a grade means what a specialist would mean by it.
Versioned methods and results
Each benchmark ships with its method, version history, and results, so scores stay comparable and reproducible as models improve.
Long-horizon, agent-oriented
Tasks span hours to days of expert work and are built for agents — multi-step, tool-using, judged end to end.
Train on real expert work.
We work directly with model post-training, agent application training, and evaluation teams. Reach out to discuss benchmark access, RL environments, or expert data.
Talk to us