Atlas

Benchmarks

← All benchmarks

RACE-Middle

General QA · 2017-04-15

RACE-Middle is the middle-school portion of RACE, a four-option multiple-choice reading-comprehension dataset built from English exams for Chinese students. Scores are accuracy on its 1,436-question test split.

Top models (higher is better)

ModelScore
Llama 3.1 405B74.2
Mixtral 8x22B73.3
Llama 3 70B73.3
DeepSeek-V273.1
DeepSeek LLM 67B Chat70.9
DeepSeek LLM 67B Base69.9
Qwen2.5 72B68.1
PaLM 540B68.1
LLaMA 65B67.9
DeepSeek-V3-Base67.1
DeepSeek LLM 7B Chat65.2
DeepSeekMoE 16B Chat64.8
PaLM 62B64.3
LLaMA 33B64.1
Qwen1.5 72B63.4
DeepSeek LLM 7B Base63.2
DeepSeekMoE 16B Base61.9
LLaMA-13B61.6
LLaMA 7B61.1
Loading Atlas data…