REQUEST SAMPLES

See graded work before you commit to volume

Pick the collections you want to see. We'll send samples in the format you'd run them in, with the grading attached.

what you can ask for

SWE-bench style

Task sets and full traces for reasoning, bug fixing, and code repair.

Terminal and agentic

Coding environments that capture tool use, shell interaction, and real execution feedback.

TAU-bench

Multi-domain evaluation for long-context reasoning and adaptive problem solving.

Cross-model evals

Standardized scoring to compare reasoning across models, releases, and fine-tuning runs.

Front-end and UI

UI generation tasks that test whether a model turns a prompt or a spec into a working interface.

Custom

Task collections and evaluation suites co-designed with your research team.

request samples