Epoch AI 能力指数

Epoch AI Capabilities Index

把多套公开基准的通过率拟合成同一条能力尺度(ECI),分数越高代表综合能力越强,可跨模型比较相对位置。

收录成绩 270 个模型 ECI 分 · 数据日期 2026-10-28 23:58

本页收录 270 个模型的ECI 分成绩,榜首是 Claude Opus 5.5(168.83), 榜内平均 132.5。 这份数据的时间是 2026-10-28 23:58(北京时间)。

收录条数 270 各榜官方公布,条条可回查
主指标 ECI 分 把多套公开基准的通过率拟合成同一条能力尺度(ECI),分数越高代表综合能力越强,可跨模型比较相对位置。
平均分 132.5 同榜内直接可比
数据日期 2026-10-28 23:58 这一批成绩的公布时间

ECI 分 排名

Epoch AI · 270 个模型

名次与分数均取自官方榜单,与官方一致。条形长度按当前这些成绩里的最大值算,领跑那条填满整列,其余按比例缩短(灰斜纹是「额外真实缺陷」,只与同列比、不与分数比)。官方没给的数据显示「—」。

Epoch AI 能力指数 ECI 分排名
名次模型95% 区间(±)ECI 分说明
1 Claude Opus 5.5 4.3 168.83 2026-09-22
2 GPT-6 Astra 4.5 168.81 2026-09-03
3 GPT-6.1 Sol 4.7 168.37 2026-09-29
4 Claude Fable 5.1 3.9 166.49 2026-09-01
5 Gemini 4 Argon 4.1 166.45 2026-09-30
6 Claude Sonnet 5.5 4.1 166.18 2026-09-28
7 Claude Opus 5 3.6 163.94 2026-07-24
8 GPT-6 Sol 3.3 163.73 2026-09-22
9 Claude Fable 5 3.8 163.42 2026-06-09
10 GPT-5.5 Pro 3.3 162.91 2026-04-23
11 GPT-5.6 Sol 3.5 162.79 2026-07-09
12 Muse Wonder 3 161.85 2026-10-12
13 GPT-5.6 Terra 3 160.09 2026-07-09
14 GPT-5.5 3.2 159.68 2026-04-23
15 GPT-5.4 Pro 2.8 159.57 2026-03-05
16 Claude Haiku 5.5 3.1 159.32 2026-10-07
17 Claude Opus 4.8 2.5 159.01 2026-05-28
18 Gemini 3.7 Flash 2.7 157.97 2026-08-13
19 Kimi K3 2.8 157.86 2026-07-16
20 Gemini 3.8 Flash 3 157.44 2026-09-02
21 Qwen 3.8 Max 2.5 157.4 2026-08-02
22 GPT-5.4 2.2 157.31 2026-03-05
23 Muse Spark 1.3 2.3 157.21 2026-09-02
24 GPT-5.3 Codex 3.9 157.09 2026-02-05
25 Claude Sonnet 5 2.8 156.88 2026-06-30
26 Claude Opus 4.7 2.1 156.79 2026-04-16
27 GPT-6 Luna 2.6 156.75 2026-09-22
28 Grok 4.6 2.3 156.57 2026-08-12
29 GPT-5.6 Luna 2.5 156.5 2026-07-09
30 GPT-5.2 Pro 3 156.33 2025-12-11
31 GLM-5.3 2.1 155.99 2026-08-14
32 DeepSeek V4.1 Flash 2.5 155.93 2026-09-09
33 Claude Opus 4.6 2.1 155.66 2026-02-05
34 DeepSeek V4 Pro 0813 2.1 155.64 2026-08-13
35 Gemini 3.1 Pro 2.4 155.34 2026-02-19
36 Muse Spark Open 2.7 155.11 2026-10-15
37 Qwen3.8 Max (0902) 2.1 155.08 2026-09-01
38 Muse Spark 1.2 2.4 155.05 2026-08-05
39 Gemini 3.5 Flash 2.2 154.88 2026-05-19
40 DeepSeek V4 Flash 0731 2.4 154.76 2026-07-31
41 Gemini 3.6 Flash 1.9 154.53 2026-07-21
42 Grok 4.7 2.2 154.37 2026-09-21
43 Grok 4.5 1.9 154.07 2026-07-08
44 Muse Spark 1.1 2.2 154.05 2026-07-09
45 Qwen3.7-Max 2.1 154.04 2026-05-19
46 GPT-5.2 1.7 153.66 2025-12-11
47 GLM-5.3-Flash 2.4 153.08 2026-08-20
48 Gemini 3 Pro 2.2 153.07 2025-11-18
49 Claude Sonnet 4.6 2.4 152.79 2026-02-17
50 Grok 4.20 2.5 152.11 2026-02-17
51 Muse Spark 2.4 152.03 2026-04-08
52 GLM-5.2 2.1 151.94 2026-06-16
53 Gemini 3 Flash 1.6 151.88 2025-12-17
54 Qwen 3.8 27B 2 151.69 2026-08-14
55 Kimi K2.6 1.8 150.94 2026-04-20
56 GPT-5 Pro 2.6 150.59 2025-10-07
57 Inkling-Small 2.4 150.32 2026-07-15
58 Kimi K2.7 Code 1.8 150.11 2026-06-12
59 Claude Opus 4.5 2.6 150.11 2025-11-24
60 Mistral Large 4 2.9 150.03 2026-10-06
61 GPT-5 — 150 2025-08-07
62 GLM-5.1 2.1 149.82 2026-04-07
63 GPT-5.1 1.7 149.76 2025-11-13
64 Qwen 3.6 Max (Preview) 2.8 149.41 2026-04-20
65 DeepSeek-V4-Pro 1.6 149.13 2026-04-24
66 Grok 4.3 Beta 1.6 148.81 2026-04-17
67 GPT-5.4 Mini 1.8 148.7 2026-03-17
68 Inkling 2.1 148.41 2026-07-15
69 Kimi K2.5 1.5 148.07 2026-01-27
70 o3-pro 2.3 147.33 2025-06-10
71 Qwen 3.6 Plus 2.1 147.3 2026-03-31
72 Qwen3.7-Plus 1.8 147 2026-06-02
73 MiniMax-M3 3.9 146.87 2026-06-01
74 Qwen3.5 397B-A17B 1.8 146.8 2026-02-13
75 Claude Sonnet 4.5 1.6 146.74 2025-09-29
76 o3 1.8 146.73 2025-04-16
77 MiniMax-M2.5 2.8 146.59 2026-02-12
78 Grok 4 1.8 146.24 2025-07-09
79 Qwen3.6 27B 1.8 146.23 2026-04-22
80 Qwen 3.5 Plus (hosted 397B-A17B) 2.2 146.17 2026-02-16
81 DeepSeek-V4-Flash 2.1 146.11 2026-04-24
82 DeepSeek-V3.2 1.6 146.09 2025-12-01
83 GPT-5.4 Nano 2.3 146.03 2026-03-17
84 Nemotron 3 Ultra 2.3 145.97 2026-06-04
85 Kimi K2 Thinking 2.1 145.82 2025-11-06
86 GLM-5 1.9 145.8 2026-02-11
87 MiniMax-M2.7 4.9 145.78 2026-03-18
88 o4-mini 2.2 145.46 2025-04-16
89 GPT-5 mini 1.9 145.34 2025-08-07
90 Gemini 2.5 Pro (Jun 2025) 1.6 145.13 2025-06-05
91 DeepSeek-V3.2-Exp 2.2 144.91 2025-09-29
92 Gemini 3.5 Flash-Lite 1.9 144.91 2026-07-21
93 Gemini 3.1 Flash-Lite 1.7 144.26 2026-03-03
94 Grok 4 Fast 2.3 144.09 2025-09-19
95 Claude Opus 4.1 2.4 144.07 2025-08-05
96 Qwen3.7 Flash 2 144.05 2026-07-27
97 Gemini 2.5 Pro (Mar 2025) 2.5 143.96 2025-03-31
98 Qwen3-235B-A22B-Thinking (Jul 2025) 1.8 143.76 2025-07-25
99 Qwen 3.6 35B-A3B 2.4 143.73 2026-04-14
100 GLM-4.7 2.2 143.59 2025-12-22
101 Qwen 3.5 Flash (hosted 35B-A3B) 1.8 143.52 2026-02-25
102 Gemma 4 31B IT 2.7 142.99 2026-04-02
103 Gemini 2.5 Flash (Sep 2025) 3.5 142.9 2025-09-25
104 Qwen 3.6 Flash 2.1 142.89 2026-04-27
105 Claude Opus 4 2 142.52 2025-05-22
106 Qwen3.5-35B-A3B 2.3 142.49 2026-02-24
107 GPT-5.5 Instant 2.8 142.43 2026-05-05
108 Qwen3-Max 2.5 142.33 2025-09-24
109 Claude Haiku 4.5 2.3 142.15 2025-10-15
110 Gemini 2.5 Pro (May 2025) 3.4 142.15 2025-05-06
111 Gemma 4 26B A4B 2.3 141.88 2026-04-02
112 o1 1.6 141.72 2024-12-17
113 Claude Sonnet 4 2 141.56 2025-05-22
114 Gemini 2.5 Flash (May 2025) 1.8 141.43 2025-05-20
115 Mistral Medium 3.5 2.9 141.36 2026-04-28
116 DeepSeek-R1 (May 2025) 1.8 141.13 2025-05-28
117 Claude 3.7 Sonnet 1.9 141.03 2025-02-24
118 Gemini 2.5 Flash (Jun 2025) 1.9 140.59 2025-06-17
119 o3-mini 2.4 140.2 2025-01-31
120 Grok-3 mini 2.1 140.2 2025-06-24
121 Kimi K2 (Jul 2025) 2.8 139.97 2025-07-12
122 Gemini 2.5 Flash (Apr 2025) 3 139.84 2025-04-17
123 gpt-oss-120b 3.3 139.79 2025-08-05
124 DeepSeek-V3.1 3.8 139.79 2025-08-21
125 Qwen3-30B-A3B-Thinking (Jul 2025) 2.8 139.55 2025-07-30
126 Qwen3.5-9B 2.4 139.32 2026-02-24
127 GPT-5 nano 3.5 139.21 2025-08-07
128 Qwen3-235B-A22B 3 139.15 2025-04-28
129 DeepSeek-R1 2.1 138.82 2025-01-20
130 Qwen3-235B-A22B-Instruct (Jul 2025) 2.7 138.78 2025-07-25
131 Qwen3-32B 2.8 138.42 2025-04-29
132 Grok 3 2 138.21 2025-04-09
133 Qwen3-14B 3.4 138.15 2025-04-29
134 gpt-oss-20b 3.7 137.7 2025-08-05
135 QwQ-32B 4.6 137.5 2025-03-05
136 Qwen3-30B-A3B-Instruct (Jul 2025) 3.8 137.27 2025-07-29
137 GPT-4.5 2.5 136.65 2025-02-27
138 GPT-4.1 2.4 136.65 2025-04-14
139 Qwen3-30B-A3B 3.9 136.11 2025-04-29
140 Qwen3-8B 4.2 136.06 2025-04-28
141 DeepSeek-V3 (Mar 2025) 2.8 135.85 2025-03-24
142 o1-mini 2.4 135.7 2024-09-12
143 Gemini 2.0 Flash Thinking (Jan 2025) 3.7 135.24 2025-01-21
144 Gemini 2.0 Pro 2.7 134.95 2025-02-05
145 GPT-4.1 mini 2.8 134.89 2025-04-14
146 Gemini 2.0 Flash (Feb 2025) 2.7 134.62 2025-02-05
147 Gemini 2.0 Flash (Dec 2024) 6.1 134.58 2024-12-11
148 o1-preview 3.6 134.57 2024-09-12
149 Mistral Medium 3 2.5 134.02 2025-05-07
150 Gemini 2.5 Flash-Lite (Jun 2025) 2.9 133.82 2025-06-17
151 Claude 3.5 Sonnet (October 2024) 4.3 133.43 2024-10-22
152 Magistral Small 1.0 2.7 133.07 2025-06-10
153 Qwen2.5-Max 3.5 132.48 2025-01-25
154 DeepSeek-V3 3.6 132.3 2024-12-26
155 Llama 4 Maverick 3.3 132.22 2025-04-06
156 Gemini 1.5 Pro (Sept 2024) 2.9 131.72 2024-09-24
157 Phi-4 3.7 130.48 2024-12-12
158 Grok-2 (Dec 2024) 3.2 130.47 2024-12-12
159 Claude 3.5 Sonnet — 130 2024-06-20
160 Gemma 3 27B 3.8 129.97 2025-03-12
161 Llama 4 Scout 3 129.63 2025-04-05
162 GPT-4.1 nano 4.2 129.54 2025-04-14
163 Gemini 1.5 Flash (Sep 2024) 3.7 129.33 2024-09-24
164 GPT-4o (May 2024) 3.3 128.99 2024-05-13
165 Qwen2.5-72B 3.6 128.97 2024-09-19
166 GPT-4o (Aug 2024) 3.4 128.83 2024-08-06
167 Llama 3.1-405B 3.1 128.77 2024-07-23
168 GPT-4o (Nov 2024) 3.5 128.69 2024-11-20
169 Mistral Large 2 (Nov 2024) 3.3 128.57 2024-11-18
170 Mistral Large 2 (Jul 2024) 3.8 127.57 2024-07-24
171 Mistral Small 3.1 3.5 127.53 2025-03-17
172 Llama 3.3 70B 3.4 127.33 2024-12-06
173 GPT-4 Turbo (Apr 2024) 3.2 127.27 2024-04-09
174 Claude 3.5 Haiku 4.4 127.07 2024-10-22
175 Gemini 1.5 Pro (May 2024) 3.5 126.94 2024-05-14
176 Claude 3 Opus 3.8 126.9 2024-02-29
177 GPT-4o mini 4.2 126.46 2024-07-18
178 GPT-4 Turbo (Nov 2023) 3.6 126.43 2024-01-25
179 Llama 3.1-70B 3.8 125.93 2024-07-23
180 GPT-4 (Mar 2023) 5.2 125.74 2023-03-14
181 Llama 3.2 90B 3.5 125.52 2024-09-24
182 Qwen2-72B 3.9 125.28 2024-06-07
183 DeepSeek-V2 (MoE-236B, May 2024) 4.5 124.8 2024-05-07
184 Gemma 3 12B 7.3 123.54 2025-03-12
185 GPT-4 (Jun 2023) 5 123.09 2023-06-13
186 Llama 3-70B 4.2 123.01 2024-04-18
187 Gemini 1.5 Flash (May 2024) 4.2 122.68 2024-05-23
188 Mistral Large 5 122.12 2024-02-26
189 Gemma 2 27B 4.3 122.12 2024-06-24
190 Mixtral 8x22B 4.6 122.04 2024-04-17
191 phi-3-small 7.4B 5.1 121.82 2024-04-23
192 phi-3-medium 14B 5 121.19 2024-04-23
193 Claude 3 Sonnet 5.3 120.84 2024-02-29
194 Claude Instant 5.2 120.21 2023-08-09
195 Claude 2 5.2 120.19 2023-07-11
196 Gemma 2 9B 5.2 119.81 2024-06-24
197 Qwen2.5-Coder-32B 7 119.45 2024-09-18
198 Claude 2.1 5 119.32 2023-11-21
199 Command R+ 7.4 119.2 2024-08-30
200 Claude 3 Haiku 5.5 118.85 2024-03-07
201 Mistral NeMo 5.1 118.65 2024-07-18
202 Qwen2.5-7B 6.1 118.54 2024-09-19
203 GPT-3.5 Turbo (Nov 2023) 5.4 118.53 2023-11-06
204 Mixtral 8x7B 4.8 118.44 2023-12-11
205 Ministral 3B 7.8 118.04 2024-10-16
206 Yi-34B 5.8 117.33 2023-11-02
207 phi-3-mini 3.8B 5.6 117.28 2024-04-23
208 Gemini 1.0 Pro 5.3 117.14 2023-12-13
209 Stable Beluga 2 5.3 117.02 2023-07-20
210 Llama 3.1-8B 7.9 116.47 2024-07-23
211 Llama 3-8B 5.9 116.39 2024-04-18
212 Qwen2.5-Coder-14B 7.9 116.31 2024-09-18
213 Gemma 3 4B 13.1 115.88 2025-03-12
214 GPT-3.5 Turbo (Jan 2024) 5.7 115.68 2024-01-25
215 PaLM 2-L 10.9 114.9 2023-05-17
216 Llama 2-70B 6.3 113.67 2023-07-18
217 GPT-3.5 Turbo (Jun 2023) 6.2 113.22 2023-06-13
218 Qwen2.5-Coder (7B) 8.1 113.01 2024-09-18
219 Qwen-14B 6.6 112.88 2023-09-24
220 Mistral 7B v0.1 6.4 112.05 2023-09-27
221 Falcon-180B 8.7 111.96 2023-09-06
222 internlm-20b 7 111.92 2023-09-18
223 Gemma 7B 6.8 111.82 2024-02-21
224 LLaMA-65B 6.8 109.97 2023-02-24
225 Falcon 2 11B 7.6 109.36 2024-05-09
226 Mistral 7B v0.3 7.2 108.88 2024-05-27
227 DeepSeek-Coder-V2-Lite-Base 7.8 108.68 2024-06-13
228 PaLM 2-M 8.5 108.06 2023-05-17
229 Phi-2 13.6 107.73 2023-12-12
230 Qwen2.5-Coder-3B 10.8 107.52 2024-09-18
231 Nemotron-4 15B 7.7 107.49 2024-02-26
232 Yi-9B 8.6 107.39 2024-03-01
233 LLaMA-33B 7.4 107.18 2023-02-24
234 Qwen-7B 8.3 106.59 2023-09-28
235 PaLM 2-S 9.2 105.91 2023-05-17
236 Llama 2-13B 8 105.58 2023-07-18
237 Llama 2-34B 7.8 104.93 2023-07-18
238 StarCoder 2 15B 10.2 104.74 2024-02-20
239 Yi 6B 8.2 104.5 2023-11-22
240 Falcon-40B 8.1 104.17 2023-05-25
241 Baichuan2-13B 9.1 102.9 2023-09-06
242 Qwen2.5-Coder (1.5B) 12.7 102.71 2024-09-18
243 internlm-7b 9 102.58 2023-07-05
244 INTELLECT-1 8.6 100.54 2024-11-29
245 MPT-30B 8.4 100.33 2023-06-22
246 LLaMA-13B 8.7 100.32 2023-02-24
247 Llama 2-7B 9.1 98.61 2023-07-18
248 chatglm2-6b 13.1 98.59 2023-06-24
249 LLaMA-7B 9.9 96.62 2023-02-24
250 Baichuan 2-7B 10.6 95.94 2023-09-20
251 DeepSeek Coder 33B 10.3 95.94 2023-11-02
252 Falcon-7B 9.9 94.7 2023-04-24
253 CodeQwen1.5-7B 11.9 94.46 2024-04-15
254 vicuna-13b-v1.1 10.3 94.25 2023-04-12
255 MPT-7B 10.6 94.19 2023-05-05
256 Gemma 2B 9.8 93.79 2024-02-21
257 StarCoder 2 7B 13 93.14 2024-02-20
258 XGen-7B 10.1 92.98 2023-06-27
259 Qwen-1_8B 15.3 92.5 2023-11-30
260 open_llama_7b 10.4 91.2 2023-06-07
261 Phi-1.5 18 91.06 2023-09-11
262 Baichuan1-7B 11.4 90 2023-06-01
263 RedPajama-INCITE-7B-Base 11.7 89.96 2023-05-04
264 Dolly 2.0-12b 12.7 89.2 2023-04-11
265 DeepSeek Coder 6.7B 12.9 89.13 2023-11-02
266 StarCoder 2 3B 12.5 88.26 2024-02-22
267 Qwen2.5-Coder-0.5B 19.6 87.63 2024-09-18
268 Cerebras-GPT-13B 12.5 82.66 2023-03-20
269 DeepSeek Coder 1.3B 16.9 62.59 2023-11-02
270 stablelm-tuned-alpha-7b 21.4 54.81 2023-04-19

评分方式与读法

这份成绩怎么看
回到官方榜核对最稳
收录成绩270 个模型分数与名次按该榜官方公布的结果列出,同一模型的不同推理档位各占一行
主指标ECI 分把多套公开基准的通过率拟合成同一条能力尺度(ECI),分数越高代表综合能力越强,可跨模型比较相对位置。
成绩时间2026-10-28 23:58这份成绩的日期(北京时间);官方不更新时这里也会停住
同榜内怎么看看分档比看名次稳相邻名次常常只差统计噪声,差 1–2 分不宜直接判定优劣
评分方式按官方原样列出,以官方榜为准官方调整评分方式时这里跟着变,旧成绩不作为结论
横向比模型配合模型智商榜一起看跨厂商综合分与降智趋势见 /model-iq,基准本身的说明见 /glossary

ECI 是拟合出来的相对指数,不是通过率,不能和各基准的百分比直接换算。 区间(±)是置信范围,重叠说明两者差距不显著。

其它权威榜单

6 个
全部榜单 →

成绩来自 Epoch AI 官方公开榜单,数据日期 2026-10-28 23:58。