Overview

评估多步骤软推理能力，包含逻辑、数学和高效推理三个领域各250个问题

Metrics

Metric	Unit	Direction
accuracy	%	↑ Higher is better

Sources

Model Score Ranking

#	Model	Vendor	Score
1	Claude 3.5 Sonnet (2024-10-22)	anthropic	71.8
2	o1	openai	71.8
3	Gemini 1.5 Pro 002	google	71.6
4	Gemini 2.0 Flash Thinking	google	70.8
5	Sonar Reasoning	other	69.4
6	Gemini 2.0 Flash	google	69.3
7	Qwen2 72B	alibaba	68.9
8	GPT-4 0125 Preview	openai	65.8
9	Sonar Large	other	65.6
10	GPT-4 Vision Preview	openai	64.5
11	Gemini 1.0 Ultra	google	64.3
12	Mistral Medium	mistral	64.3
13	Claude 3 Sonnet	anthropic	64.1
14	o1 Preview	openai	64.1
15	Qwen1.5 110B	alibaba	63.9
16	Yi Large Turbo	other	63.7
17	Claude 3 Sonnet (2024-02-29)	anthropic	63.5
18	GPT-4o (2024-08-06)	openai	62.9
19	GPT-4o (2024-05-13)	openai	62.8
20	Llama 3.3 70B	meta	62.3
21	Yi Vision	other	61.4
22	GPT-4 Turbo	openai	61.1
23	Gemini 1.5 Flash 002	google	59.8
24	Orca 2 13B	other	59.8
25	Yi Large	other	59.4
26	Falcon 180B	other	59.1
27	Qwen1.5 14B	alibaba	58.7
28	GPT-4 1106 Preview	openai	58.4
29	Claude 3 Haiku	anthropic	58.1
30	StableLM 2 12B	other	58.1
31	Microsoft WizardLM 2 8x22B	other	58
32	Gemini 1.0 Pro	google	57.8
33	Grok-2 Vision	xai	57.7
34	Hermes 3 Llama 3.1 405B	other	57.7
35	Zephyr ORPO 141B Alpha	other	57.7
36	Command Nightly	cohere	57.5
37	Nous Hermes 2 Mixtral 8x7B	other	57.4
38	Claude 3 Opus (2024-02-29)	anthropic	57.3
39	Jamba 1.5	other	57.1
40	Command R	cohere	57
41	Jamba Instruct	other	56.9
42	WizardLM Team WizardLM 2 8x22B	other	56.8
43	GLM-4 Flash	other	56
44	Gemini 1.0 Flash	google	55.9
45	Hermes 3 Llama 3.1 70B	other	55.4
46	Sonar Huge	other	55.4
47	GPT-4 32K	openai	54.9
48	Qwen2.5 32B	alibaba	54.9
49	DBRX Instruct	other	54.7
50	Gemini 1.5 Flash-8B 002	google	54.7
51	GPT-4	openai	54.6
52	Qwen1.5 32B	alibaba	54.5
53	Llama 3.1 Nemotron 70B	meta	54.1
54	o1 mini	openai	54
55	DBRX Base	other	53.8
56	Llama 3.2 90B Vision	meta	53.5
57	Mixtral 8x7B	mistral	53.5
58	DeepSeek V2	deepseek	53.4
59	Microsoft WizardMath 7B v1	other	53.4
60	Claude 3 Opus	anthropic	53.3
61	GPT-4o mini	openai	53.1
62	Microsoft WizardCoder Python 34B	other	53.1
63	Gemma 2 27B	google	53
64	Claude 3.5 Haiku	anthropic	52.3
65	Mistral Large	mistral	52.1
66	DeepSeek V2 Chat	deepseek	52
67	DeepSeek LLM 67B	deepseek	51.9
68	StableCode 3B	other	51.8
69	StarCoder2 15B	other	51.7
70	Grok-2 Mini	xai	51.4
71	GLM-4 Plus	other	51.3
72	GPT-4 Vision	openai	51.2
73	Llama 3 70B	meta	51.2
74	Command R+ (08-2024)	cohere	51.1
75	Jamba 1.5 Large	other	51.1
76	Claude 3 Haiku (2024-03-07)	anthropic	51
77	Yi 1.5 34B	other	51
78	Qwen1.5 72B	alibaba	50.8
79	DeepSeek Coder 7B	deepseek	50.5
80	Qwen2.5 14B	alibaba	50.5
81	Qwen2 57B	alibaba	50.5
82	Phi-3.5 MoE	other	50.3
83	Code Bison	google	49.8
84	DeepSeek Math 7B	deepseek	49.8
85	GLM-4 9B Chat	other	49.5
86	Microsoft WizardLM 2 7B	other	49.3
87	Mistral Small	mistral	49.3
88	Command R (08-2024)	cohere	49.1
89	Phi-1	other	49
90	Nous Hermes 2 Yi 34B	other	48.8
91	Jamba 1.5 Mini	other	48.1
92	Llama 3.1 70B	meta	48.1
93	Mistral 7B v0.3	mistral	48.1
94	NVIDIA Llama 3.1 Nemotron 70B	other	47.7
95	Sonar Small	other	47.6
96	Code Llama 13B	meta	47.3
97	Codestral	mistral	47.2
98	Codestral Mamba	mistral	47
99	Command R7B	cohere	47
100	Qwen2.5 7B	alibaba	46.9
101	OLMo 7B	other	46.8
102	GLM-4 Air	other	46.7
103	Yi 1.5 6B	other	46.6
104	Yi 1.5 9B	other	46.4
105	Gemini 1.5 Flash	google	46.1
106	Llama 3 8B	meta	46.1
107	Mistral Small 3	mistral	46
108	DeepSeek Coder 33B	deepseek	45.8
109	Gemini 1.5 Flash-8B	google	45.6
110	Code Llama 34B	meta	44.9
111	Grok Vision Beta	xai	44.5
112	StarChat2 15B v0.1	other	44.4
113	OLMo 7B SFT	other	44.3
114	Gemma 2 9B	google	44.1
115	Phi-3 Medium	other	43.9
116	Qwen2 7B	alibaba	43.9
117	Grok Beta	xai	43.7
118	ChatGLM3 6B	other	42.9
119	Code Llama 7B	meta	42.9
120	Jurassic-2 Ultra	other	42.9
121	Llemma 7B	other	42.6
122	OLMo 1.7 7B	other	42.6
123	Nous Capybara 7B	other	42.5
124	OLMo 7B Instruct	other	42.5
125	MPT 7B	other	42.4
126	Mathstral 7B	mistral	42.3
127	Flan-T5 XXL	other	42.2
128	Llama 3.1 8B	meta	41.7
129	Llama 3.2 11B Vision	meta	41.7
130	Mistral Tiny	mistral	41.4
131	Text Bison	google	40.8
132	Command Light	cohere	40.7
133	Zephyr 7B Alpha	other	40.5
134	GPT-3.5	openai	40.4
135	DeepSeek Coder V2	deepseek	40.3
136	Flan-T5 XL	other	40.2
137	Code Llama 70B	meta	40.1
138	Open-Platypus	other	40
139	Nous Hermes 2 Solar 10.7B	other	39.6
140	TigerBot 70B Chat	other	39.3
141	OLMo 2 1124 7B	other	39.2
142	Phi-4	other	38.6
143	GLM-4V 9B	other	38.5
144	Mistral Nemo	mistral	38.5
145	StarCoder2 3B	other	38.4
146	StarCoder2 7B	other	38.3
147	OpenChat 3.6 8B	other	38.2
148	Falcon 40B	other	38.1
149	Hermes 3 Llama 3.1 8B	other	38
150	Baichuan2 13B Chat	other	37.9
151	Qwen2.5 1.5B	alibaba	37.8
152	Claude 2.1	anthropic	37.7
153	Mistral 7B v0.1	mistral	37.4
154	Flan-UL2	other	37.2
155	Ministral 8B	mistral	37
156	Yi 34B	other	36.8
157	Mistral 7B v0.2	mistral	36.7
158	StableLM 3 4B	other	36.6
159	Phi-3 Small	other	36.5
160	Gemma 7B	google	36.2
161	Llama 2 70B	meta	36.1
162	Llama 2 7B	meta	35.3
163	Claude 2	anthropic	35.2
164	Llama Guard 2 8B	meta	34.9
165	Capybara 1.5B	other	34
166	Llama 2 13B	meta	33.8
167	Claude Instant 1	anthropic	33.5
168	Phi-3 Mini	other	33.5
169	Baichuan2 7B Chat	other	33.1
170	Chat Bison	google	32.8
171	OpenChat 3.5 1210	other	32.6
172	Zephyr 7B Beta	other	32.6
173	Jurassic-2 Mid	other	32.3
174	Phi-3 Vision	other	32.1
175	Nous Capybara 34B	other	31.7
176	Phi-3.5 Mini	other	31.2
177	Gemma 2B	google	30.6
178	Phi-1.5	other	30.3
179	Argilla Notus 7B v1	other	30.1
180	Yi 6B	other	29.5
181	Ministral 3B	mistral	29.4
182	Llama 3.2 3B	meta	29.3
183	PaLM 2	google	29.3
184	GPT-3.5 Turbo	openai	29.2
185	GPT-3.5 Turbo 16K	openai	29.1
186	MPT 30B	other	28.9
187	Qwen2.5 0.5B	alibaba	28.9
188	Qwen2 1.5B	alibaba	27.8
189	Llama 3.2 1B	meta	26.6
190	StableLM 2 1.6B	other	26.4
191	Phi-2	other	23.7
192	StableLM Zephyr 3B	other	23.5
193	Qwen2.5 3B	alibaba	23.3
194	Llama Guard 3 8B	meta	22.8
195	Embed English v3	cohere	0

MUSR

설명

评估多步骤软推理能力，包含逻辑、数学和高效推理三个领域各250个问题

핵심 사양

카테고리	라이선스	최종 업데이트
reasoning	MIT	2024-02-01

벤치마크

단위
%

소스

공식 URL

https://github.com/Zayne-sprague/MUSR