TriviaQA

TriviaQA — reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents.

knowledge

Models Tested

87.6

Top Score

75.4

Average Score

Rankings

#	Model	Score
1	Llama 2-70BMeta	87.6
2	Claude 2Anthropic	87.5
3	LLaMA-65BMeta	86.0
4	GPT-4 TurboOpenAI	84.8
5	GPT-4 Turbo (older v1106)OpenAI	84.8
6	GPT-4 (older v0314)OpenAI	84.8
7	GPT-4OpenAI	84.8
8	Llama 2-34BMeta	84.6
9	LLaMA-33BMeta	83.8
10	DeepSeek V3DeepSeek	82.9
11	Llama 3.1-405BMeta	82.7
12	Mixtral 8x7B InstructMistral AI	82.2
13	DeepSeek-V2 (MoE-236B, May 2024)DeepSeek	80.0
14	T Falcon-40BTII	79.9
15	Llama 2-13BMeta	79.6
16	Claude InstantAnthropic	78.9
17	LLaMA-13BMeta	77.9
18	Mistral 7B v0.1Mistral	75.2
19	phi-3-medium 14BMicrosoft	73.9
20	Llama 2-7BMeta	73.7
21	U MPT-30Bunknown	73.6
22	Gemma 7BGoogle	72.3
23	Qwen2.5 72B InstructAlibaba Qwen	71.9
24	LLaMA-7BMeta	71.0
25	Llama 3 8B InstructMeta	67.7
26	T Falcon-7BTII	64.6
27	phi-3-mini 3.8BMicrosoft	64.0
28	U MPT-7Bunknown	61.6
29	phi-3-small 7.4BMicrosoft	58.1
30	Gemma 2BGoogle	53.2
31	Phi-2Microsoft	45.2