mpt-7B-chat: benchmark results

Provider: Databricks. Released 2023-05-05. Access: Open.

Unified ELO 1305 ± 10, rank #2681 of 2928 rated models, from 99 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
LLM Trustworthy - Fairness100Trust Score (%)93.8
MMLU-by-task - College Mathematics36Accuracy (%)80.3
MMLU-by-task - Virology42.77Accuracy (%)60.2
MMLU-by-task - Electrical Engineering45.52Accuracy (%)56
MMLU-by-task - Machine Learning33.04Accuracy (%)53.1
MMLU-by-task - Abstract Algebra30Accuracy (%)52.1
LLM Trustworthy - Adversarial46.2Trust Score (%)50
LLM Trustworthy - Out-of-Distribution64.26Trust Score (%)50
MMLU-by-task - Security Studies48.57Accuracy (%)48.8
MMLU-by-task - College Chemistry32Accuracy (%)46.7
MMLU-by-task - High School Computer Science41Accuracy (%)46.2
LLM Trustworthy - Privacy78.93Trust Score (%)45.8

Interactive version: theaggregate.ai/model?slug=mpt-7b-chat · How It Works · Data refreshed daily, snapshot 2026-09-23.