Baichuan2-7B-Chat: benchmark results

Baichuan Intelligence's Baichuan2-7B-Chat, trained on a 2.6-trillion-token corpus of Chinese and English text, the chat version of their 7B model in the Baichuan 2 family. Provider: Baichuan. Released 2023-09-06. Access: Open.

Aggregate Rating 1334 ± 20, rank #1401 of 1635 rated models, from 94 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
OpenCompass Language - Content Summarization - Chinese (CompassBench 2404)24.1Score (%)56.7
OpenCompass Language - Translation (CompassBench 2404)14.5Score (%)56.7
OpenCompass Language - Content Summarization - English (CompassBench 2404)29.8Score (%)53.3
SuperCLUE General (September 2023) - Open-Ended Questions (OPEN)42.28Score47.4
SuperCLUE General (September 2023) - Overall50.11Score47.4
OpenCompass Knowledge - Humanities - Chinese (CompassBench 2404)34.4Score (%)46.7
OpenCompass Language - Intention Recognition - Chinese (CompassBench 2404)29.7Score (%)46.7
OpenCompass Reasoning - Common Sense - Chinese (CompassBench 2404)11.9Score (%)46.7
LLMsPark - Dictator Game (Multiple Play)989.97Elo rating (initial 1000, K=32)46.2
LLMsPark - Dictator Game (Single Play)993.92Elo rating (initial 1000, K=32)46.2
OpenCompass Knowledge - Common Sense - Chinese (CompassBench 2404)44.3Score (%)45
OpenCompass Language - Traditional Cultural Understanding - Chinese (CompassBench 2404)32Score (%)45

Interactive version: theaggregate.ai/model?slug=baichuan2-7b-chat · How It Works · Data refreshed daily, snapshot 2026-10-15.