DBRX Instruct: benchmark results

Provider: Databricks. Released 2024-03-27. Access: Open.

Unified ELO 1359 ± 1, rank #1252 of 1392 rated models, from 16 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
HELM Lite32.55Mean win rate (self-reported)26.3
HELM (Stanford)28.87Mean Win Rate (%)21.1
HELM Safety BBQ79.2BBQ accuracy (%)18.6
HELM NaturalQuestions (Closed)28.39F1 (%)15.6
HELM WMT 201413.12BLEU-4 (%)12.2
AA GPQA Diamond33.13Accuracy (%)9.7
Artificial Analysis Intelligence Index1Intelligence Index9.4
HELM NarrativeQA48.84F1 (%)7.8
HELM NaturalQuestions (Open)54.98F1 (%)5.6
HELM Safety XSTest77.4LM Evaluated Safety score (%)2.3
AA Humanity's Last Exam2.92Accuracy (%)2.2
HELM Safety62.8Mean score (self-reported)2

Interactive version: theaggregate.ai/model?slug=dbrx-instruct · How It Works · Data refreshed daily, snapshot 2026-09-05.