Independent construction study · 24 apps

RepoPrompt Discord ViBench Showcase

Review application outputs from four native-agent treatments. Construction is in progress. Evaluation is pending.

Construction showcase Evaluation pending Core package TBD

Corpus progress

Construction coverage

Verified means the output passed construction checks. It is not an evaluator score.

Sol

24 / 24

24 outputs · 24 verified

Grok

23 / 24

20 verified · 3 provisional

Fable + Opus

23 / 24

Fable 5.1 High → Opus 5 High

21 verified · 2 flagged · 1 failed

RepoPrompt CE Default

0 / 24

Not run

24-task MVP corpus

Output catalog

Verified Provisional Missing Not run

Find a task

Showing 24 tasks

Construction status and preview controls for every ViBench MVP task.
Task	Description	Pinned spec	Sol	Grok	Fable + Opus	RepoPrompt CE Default
Loading the 24-task catalog…

Method

One fixed evaluator

Each product uses its native agent tools for construction. The ViBench evaluator is fixed, independent, and uses separate API credentials. Evaluation is not complete. Construction status is not a score.

Fable 5.1 directed Opus 5 through native Claude Code agents. All 24 tasks were attempted. HVAC and Notes retain scanner flags, reported as false positives. Quiz failed during output finalization and is not available. High effort was set for both models. The CLI did not report the effective effort. These construction results have not been scored.