Terminal Bench 2.1

Roybench

Evidence-backed Terminal Bench 2.1 Full88 aggregates by exact provider route, model, reasoning effort, harness, and version. Missing or unusable evidence stays explicit.

EnvironmentDetecting…
Generated atNot reported
Data source/data/leaderboard.json
Loading leaderboard data…
Source rowsPublished benchmark records
Model variantsUnique route-qualified configurations
Best accuracyExact rankable rows
Top pass coverageTasks passed / expected
Scoring and ranking semantics

Ranked evidence

Top performers — Accuracy

Passed / 88, evidence coverage, then full lexical identity

Full comparison

Benchmark configurations

Benchmark configurations. Full88 rows follow the selected metric ordering; compatible legacy rows can be sorted by column heading. Inspect a row to view every source field.
Dataset-level provenance and fields