排行榜/Arena 人类盲测/Search 搜索

Arena Search 搜索排行榜

联网搜索型模型 Elo 排名 · 完整榜单 42 个模型

Rank	区间	Model	厂商	License	Elo 分	投票数	输入 $/M	输出 $/M	上下文
1	1–3	gpt-5.6-sol-xhigh	OpenAI	Proprietary	1261.1±6.3	60.4k	$4	$20	1M
2	1–6	qwen3.8-max	Alibaba	Proprietary	1261.1±13.2	1.9k	$1.7	$5.1	1M
3	1–6	claude-opus-4-6-search	Anthropic	Proprietary	1251.4±4.6	173.2k	$5	$25	1M
4	2–7	gpt-5.5-search	OpenAI	Proprietary	1246.6±4.9	126.0k	$5	$30	1M
5	2–8	claude-fable-5.1-high	Anthropic	Proprietary	1245.8±7.1	12.8k	$5	$25	1M
6	2–11	claude-opus-5-high	Anthropic	Proprietary	1241.6±6.9	13.3k	$5	$25	1M
7	5–11	gpt-6-astra-max	OpenAI	Proprietary	1233.6±7.9	18.5k	$10	$50	1M
8	4–12	gpt-6.1-sol-max	OpenAI	Proprietary	1233.1±8.9	5.6k	$2	$10	1M
9	6–11	claude-opus-4-7	Anthropic	Proprietary	1230.1±4.9	127.5k	$5	$25	1M
10	6–12	claude-fable-5-high	Anthropic	Proprietary	1229.1±6.5	73.6k	$10	$50	1M
11	6–14	ernie-5.1	Baidu	Proprietary	1226.7±9.5	4.0k	$0.56	$2.5	119K
12	9–14	claude-sonnet-4-6-search	Anthropic	Proprietary	1220.5±4.5	174.0k	$1.5	$7.5	1M
13	11–18	grok-4.5	SpaceXAI	Proprietary	1215.9±5.9	76.5k	$2	$6	500K
14	11–19	claude-opus-5.5-medium	Anthropic	Proprietary	1212.8±7	10.8k	$4	$20	1M
15	13–21	gemini-3.1-pro-grounding	Google	Proprietary	1208.8±4.7	151.4k	-	-	-
16	13–21	gpt-5.2-search	OpenAI	Proprietary	1207.2±5.5	55.6k	$0.88	$7	400K
17	13–21	gemini-3-pro-grounding	Google	Proprietary	1206.5±5.5	38.9k	$2	$12	-
18	13–22	grok-4.20-multi-agent-beta-0309	SpaceXAI	Proprietary	1204.8±5.4	126.2k	$1.3	$2.5	1M
19	14–22	claude-opus-4-8	Anthropic	Proprietary	1204.1±5.6	106.3k	$5	$25	1M
20	15–24	gpt-5.1-search	OpenAI	Proprietary	1199.7±5.2	63.0k	$0.63	$5	400K
21	15–24	gpt-5.4-search	OpenAI	Proprietary	1199.1±5.5	126.7k	$2.5	$15	1M
22	18–24	gemini-3-flash-grounding	Google	Proprietary	1195.7±4.3	196.4k	-	-	-
23	20–25	claude-sonnet-5-search	Anthropic	Proprietary	1190.9±6.1	73.0k	$2	$10	1M
24	20–25	grok-4.20-beta1	SpaceXAI	Proprietary	1190.5±6	57.0k	-	-	-
25	23–29	claude-opus-4-5-search	Anthropic	Proprietary	1179.6±5.6	65.0k	$5	$25	200K
26	25–30	gpt-6-sol-max	OpenAI	Proprietary	1174.9±7.1	11.3k	$1	$5	1M
27	25–31	gpt-5.2-search-non-reasoning	OpenAI	Proprietary	1171.5±5.5	79.8k	$0.88	$7	400K
28	25–31	grok-4-1-fast-search	SpaceXAI	Proprietary	1170.9±5	85.9k	$1.3	$2.5	2M
29	25–31	grok-4-fast-search	SpaceXAI	Proprietary	1170.1±4.6	43.1k	$0.20	$0.50	2M
30	26–32	grok-4.3	SpaceXAI	Proprietary	1165.8±4.7	128.4k	$1.3	$2.5	1M
31	30–32	claude-sonnet-4-5-search	Anthropic	Proprietary	1160.3±5.2	144.2k	$3	$15	1M
32	27–33	gpt-6-luna-max	OpenAI	Proprietary	1159.6±7.3	10.8k	$0.05	$0.25	1M
33	32–37	claude-opus-4-1-search	Anthropic	Proprietary	1148.4±4.9	79.8k	$15	$75	200K
34	33–38	o3-search	OpenAI	Proprietary	1144.8±5.6	20.8k	$1	$4	200K
35	33–38	gemini-2.5-pro-grounding	Google	Proprietary	1142.8±4.7	86.7k	$0.63	$5	1M
36	33–39	grok-4-search	SpaceXAI	Proprietary	1141.9±6.1	19.2k	$3	$15	-
37	33–39	ppl-sonar-reasoning-pro-high	Perplexity AI	Proprietary	1139±5.8	29.4k	$1	$1	127K
38	34–40	gpt-5-search	OpenAI	Proprietary	1133.5±6	20.9k	$0.63	$5	400K
39	36–40	ppl-sonar-pro-high	Perplexity AI	Proprietary	1130.3±5.8	28.9k	$1	$1	127K
40	38–40	claude-opus-4-search	Anthropic	Proprietary	1127.4±5.4	31.5k	$15	$75	200K
41	41–42	diffbot-small-xl	Diffbot	Apache 2.0	1022.7±8.5	6.4k	-	-	-
42	41–42	api-gpt-4o-search	OpenAI	Proprietary	1006.6±11.2	3.4k	$30	$60	8K

Elo 分真人在不知道模型名字的情况下两两投票，胜负累积出的等级分。绝对值没有意义，差值才有：分差 100 分 ≈ 强的那个有 64% 胜率，只差几分基本是平手。
± 后面的数是 95% 置信区间的半宽，投票越多越小。两个模型的区间重叠，说明它们的名次差别在统计上不成立。
区间把上面这个置信区间换算成名次：1–5 意思是这个模型的真实名次可能在第 1 到第 5 之间。

数据更新: 2026-10-14数据来源: Arena AI ↗