Triplet Incremental Training Study - NuInstruct Accuracy: leaderboard

Triplet Incremental Training v1: unchanged zero-shot InternVL3.5-4B on the source native driving-language panel; exact inference prompt/toolkit/score normalization not reported. No SFT Baseline/Ours score is assigned to this checkpoint. Source singleton measurements remain evidence; no ranking eligibility is inferred. Metric: Native accuracy (%) Higher is better.

Metric: Native accuracy (%). Source: arxiv.org. 1 model tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-4B0.34#1916

Interactive version: theaggregate.ai/benchmark?slug=triplet-incremental-training-study-nuinstruct-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-16.