Benchmarks / Composite

Epoch Capabilities Index

Epoch AI's Capabilities Index: a single score per model fitted across many benchmarks (item-response model), so scores are comparable across models tested on different sets. About 100 is GPT-4.1 and about 150 is a current frontier model. Data: Epoch AI, CC BY.

184models

–Open Index weight

–last updated

epoch-ecipublisher

#	Model	Effort	Score	Cost / task	Harness
1	Opus 5.5	max	167	–	–
2	GPT-6 Astra	max	166	–	–
3	GPT-6.1 Sol	max	166	–	–
4	Fable 5.1	max	165	–	–
5	Gemini 4	max	165	–	–
6	Sonnet 5.5	max	164	–	–
7	Opus 5	max	163	–	–
8	GPT-6 Sol	max	163	–	–
9	Fable 5	max	162	–	–
10	GPT-5.5 Pro	xhigh	162	–	–
11	GPT-5.6 Sol	max	162	–	–
12	GPT-5.6 Terra	max	159	–	–
13	GPT-5.4 Pro	xhigh	159	–	–
14	Haiku 5.5	max	159	–	–
15	GPT-5.5	xhigh	159	–	–
16	Opus 4.8	max	158	–	–
17	Kimi K3	max	157	–	–
18	Gemini 3.7 Flash	max	157	–	–
19	Gemini 3.8 Flash	max	157	–	–
20	Qwen 3.8 Max	max	157	–	–
21	GPT-5.3 Codex	xhigh	157	–	–
22	Muse Spark 1.3	max	157	–	–
23	GPT-5.4	xhigh	157	–	–
24	Sonnet 5	max	156	–	–
25	GPT-5.6 Luna	max	156	–	–
26	Opus 4.7	max	156	–	–
27	GPT-6 Luna	max	156	–	–
28	Grok 4.6	max	156	–	–
29	GPT-5.2 Pro	xhigh	156	–	–
30	GLM 5.3	max	156	–	–
31	DeepSeek V4.1 Flash	max	156	–	–
32	DeepSeek V4 Pro 0813	max	155	–	–
33	Opus 4.6	max	155	–	–
34	Muse Spark 1.2	max	155	–	–
35	Gemini 3.1 Pro	high	155	–	–
36	DeepSeek V4 Flash 0731	max	154	–	–
37	Gemini 3.5 Flash	xhigh	154	–	–
38	Grok 4.7	max	154	–	–
39	Gemini 3.6 Flash	xhigh	154	–	–
40	Muse Spark 1.1	max	154	–	–
41	Grok 4.5	max	154	–	–
42	Qwen 3.7 Max	max	154	–	–
43	GPT-5.2	xhigh	153	–	–
44	GLM 5.3 Flash	max	153	–	–
45	Gemini 3 Pro	high	153	–	–
46	Sonnet 4.6	max	152	–	–
47	Muse Spark	–	152	–	–
48	Grok 4.20	xhigh	152	–	–
49	GLM 5.2	max	152	–	–
50	Gemini 3 Flash	high	152	–	–
51	Qwen 3.8 27B	max	152	–	–
52	Kimi K2.6	max	151	–	–
53	GPT-5 Pro	high	151	–	–
54	Inkling Small	xhigh	150	–	–
55	Opus 4.5	high	150	–	–
56	GPT-5	high	150	–	–
57	GLM 5.1	max	150	–	–
58	Kimi K2.7 Code	max	150	–	–
59	GPT-5.1	high	150	–	–
60	Qwen 3.6 Max	max	149	–	–
61	DeepSeek V4 Pro	max	149	–	–
62	Grok 4.3	xhigh	149	–	–
63	GPT-5.4 Mini	xhigh	149	–	–
64	Inkling	max	149	–	–
65	Kimi K2.5	max	148	–	–
66	Qwen 3.6 Plus	max	147	–	–
67	o3 pro	high	147	–	–
68	Qwen 3.7 Plus	high	147	–	–
69	MiniMax M3	max	147	–	–
70	Mistral Large 4	high	147	–	–
71	o3	high	147	–	–
72	Sonnet 4.5	high	147	–	–
73	MiniMax M2.5	high	147	–	–
74	Qwen 3.5 397B-A17B	max	147	–	–
75	Qwen 3.6 27B	max	146	–	–
76	Grok 4	high	146	–	–
77	Qwen 3.5 Plus	none	146	–	–
78	GPT-5.4 Nano	xhigh	146	–	–
79	DeepSeek V3.2	high	146	–	–
80	DeepSeek V4 Flash	max	146	–	–
81	Kimi K2	medium	146	–	–
82	MiniMax M2.7	high	146	–	–
83	GLM 5	high	146	–	–
84	Nemotron 3 Ultra 550B A55B	max	146	–	–
85	o4 mini	high	146	–	–
86	GPT-5 Mini	high	145	–	–
87	Gemini 2.5 Pro	high	145	–	–
88	Gemini 3.5 Flash-Lite	xhigh	145	–	–
89	Qwen 3.7 Flash	max	145	–	–
90	Gemini 3.1 Flash-Lite	high	144	–	–
91	Grok 4 Fast	none	144	–	–
92	Qwen 3.6 35B-A3B	max	144	–	–
93	Opus 4.1	high	144	–	–
94	Qwen 3 235B-A22B	xhigh	144	–	–
95	GLM 4.7	none	144	–	–
96	Qwen 3.5 Flash	max	144	–	–
97	Qwen 3.6 Flash	max	143	–	–
98	Gemini 2.5 Flash	medium	143	–	–
99	Gemma 4 31B IT	high	143	–	–
100	Opus 4	medium	143	–	–
101	Qwen 3.5 35B-A3B	low	142	–	–
102	GPT-5.5 Instant	–	142	–	–
103	Haiku 4.5	xhigh	142	–	–
104	Qwen 3 Max	max	142	–	–
105	o1	high	142	–	–
106	Gemma 4 26B	xhigh	142	–	–
107	Sonnet 4	high	142	–	–
108	Mistral Medium 3.5	high	141	–	–
109	DeepSeek R1	xhigh	141	–	–
110	Sonnet 3.7	medium	141	–	–
111	o3 mini	high	140	–	–
112	Grok 3 Mini	high	140	–	–
113	gpt-oss-120b	high	140	–	–
114	DeepSeek V3.1	none	140	–	–
115	Qwen 3 30B-A3B	none	140	–	–
116	Qwen 3.5 9B	low	139	–	–
117	GPT-5 Nano	high	139	–	–
118	Qwen 3 32B	none	138	–	–
119	Grok 3	none	138	–	–
120	Qwen 3 14B	none	138	–	–
121	gpt-oss-20b	high	138	–	–
122	GPT-4.1	high	137	–	–
123	GPT-4.5	none	137	–	–
124	Qwen 3 8B	none	136	–	–
125	DeepSeek V3	none	136	–	–
126	o1 mini	high	136	–	–
127	Gemini 2.0 Flash	medium	135	–	–
128	Gemini 2.0 Pro	none	135	–	–
129	GPT-4.1 Mini	high	135	–	–
130	Mistral Medium 3	none	134	–	–
131	Gemini 2.5 Flash Lite	none	134	–	–
132	Magistral Small 1.0	–	133	–	–
133	Qwen 2.5 Max	none	133	–	–
134	Llama 4 Maverick	medium	132	–	–
135	Mistral Small 3.2	none	132	–	–
136	Gemini 1.5 Pro	none	132	–	–
137	Magistral Small 1.2	–	131	–	–
138	Grok 2	none	130	–	–
139	Gemma 3 27B	none	130	–	–
140	Llama 4 Scout	medium	130	–	–
141	GPT-4.1 Nano	high	130	–	–
142	Gemini 1.5 Flash	none	129	–	–
143	GPT-4o	high	129	–	–
144	Qwen 2.5 72B	none	129	–	–
145	Llama 3.1 405b	none	129	–	–
146	Qwen 2.5 32B	none	129	–	–
147	Mistral Large 2	none	129	–	–
148	Mistral Small 3.1	none	127	–	–
149	Llama 3.3 70b	none	127	–	–
150	GPT-4	high	127	–	–
151	Haiku 3.5	medium	127	–	–
152	Mistral Small 3	none	127	–	–
153	Opus 3	none	127	–	–
154	GPT-4o mini	high	127	–	–
155	Llama 3.1 70b	none	126	–	–
156	Llama 3.2 90b	none	126	–	–
157	Qwen 2 72B	none	125	–	–
158	DeepSeek V2	none	125	–	–
159	Gemma 3 12B	none	123	–	–
160	Llama 3 70b	none	123	–	–
161	Gemma 2 27B	–	122	–	–
162	Mistral Large	none	122	–	–
163	Sonnet 3	none	121	–	–
164	Gemma 2 9B	–	120	–	–
165	Qwen 2.5 Coder	none	120	–	–
166	Command-R+ (Apr '24)	none	119	–	–
167	Mistral	none	119	–	–
168	GPT-3.5	high	119	–	–
169	Qwen 2.5 7B	–	119	–	–
170	Mixtral 8x7B Instruct	none	119	–	–
171	Haiku 3	medium	118	–	–
172	Gemini 1.0 Pro	none	117	–	–
173	Llama 3.1 8b	medium	117	–	–
174	Llama 3 8b	none	116	–	–
175	Gemma 3 4B	none	116	–	–
176	Llama 2 70b	none	114	–	–
177	Llama 6 5b	none	110	–	–
178	Llama 3 3b	–	107	–	–
179	Llama 2 13b	none	106	–	–
180	Llama 2 34b	–	105	–	–
181	Llama 3.2 1b	none	102	–	–
182	Llama 1 3b	–	101	–	–
183	Llama 2 7b	none	99	–	–
184	Qwen 1 8B	–	93	–	–