Morrow Labby Topify AI

Benchmark-facing data

Coding benchmark catalog.

Designed by a core author of Terminal-Bench.

Name

Terminal Bench Science

Description

Challenging, verifier-backed environments covering diverse topics across scientific computing, engineering, systems, and applied research.

Supply

Capacity 200 / week

Name

Terminal Bench 2.1

Description

Containerized tasks with reproducible state, executable verifiers, and outcome-based scoring.

Supply

OTS 2,000

Capacity 500 / week

Name

Terminal Bench 3

Description

Verifier-backed environments targeting compound systems and software-engineering failure modes.

Supply

OTS 500

Capacity 500 / week

Availability

Coming soon

Name

Auto Research Kernel Optimization Tasks

Description

Executable GPU-kernel optimization environments with correctness gates and latency and memory objectives.

Supply

OTS 2,400

Capacity 1,000 / week

Name

Long-Tail Systems Coding Benchmark

Description

Containerized systems tasks across non-mainstream languages and toolchains, with executable acceptance tests and repository state.

Supply

Capacity 200 / week

Name

Complex Office Agent Benchmark

Description

Stateful cross-application workflows requiring evidence reconciliation, artifact generation, and rubric-based evaluation.

Supply

Capacity 200 / week

Name

Advanced Spreadsheet and Audit Benchmark

Description

Workbook-scale tasks covering formula repair, audit sampling, statistical checks, and traceable artifact validation.

Supply

Capacity 300 / week

Name

Long-Horizon Task — Large Codebase

Description

Repository-scale environments for cross-module implementation, regression testing, and patch-level evaluation.

Supply

Capacity 100 / week

Name

Long-Horizon Task — Coding / Working

Description

Persistent coding and computer-use environments with multi-stage dependencies, tool calls, failure recovery, and end-state verification.

Supply

Capacity 200 / week

Name

Multi-Audience Daily Coordination Benchmark

Description

Cross-application coordination tasks evaluated through evidence consistency, audience-specific artifacts, and traceable completion.

Supply

Capacity 400 / week

Name

SWE-bench

Description

Repository issue-resolution environments with reproducible builds, test harnesses, and patch-level grading.

Supply

OTS 2,000

Capacity 1,000 / week

Availability

Coming soon

Need a specific coding dataset?

Request the matching data card and sample.

Specify the target capability, environment interface, and evaluation protocol.

Contact the data team