AI Arena Rankings (Elo)

Community arena Elo for 40 AI models — where human voters, not benchmarks, decide the winner.

Snapshot 2026-10-14 · updated weekly

Arena rankings come from blind head-to-head votes: a human picks the better of two anonymous model answers, and wins feed a Bradley-Terry / Elo rating. This page mirrors the community arena consensus each week alongside our benchmark-based AMR aggregate, so you can see where human preference and benchmarks disagree. The Elo values below are compiled from public community arena results; vote counts and confidence intervals are on each model page.

Rank	Model	Vendor	Arena Elo	AMR Score
1	Claude Opus 5.5	Anthropic	1489	96.8
2	GPT-6.1 Sol	OpenAI	1478	95.9
3	GPT-6 Astra	OpenAI	1474	95.6
4	Claude Fable 5.1	Anthropic	1466	94.9
5	Claude Sonnet 5.5	Anthropic	1462	94.7
6	Claude Fable 5	Anthropic	1455	93.8
7	Gemini 4 Argon	Google	1451	93.2
8	Claude Opus 5	Anthropic	1447	92.6
9	GPT-6 Sol	OpenAI	1440	91.4
10	GPT-5.6 Sol	OpenAI	1428	89.7
11	Gemini 4.0	Google	1421	89.0
12	DeepSeek Harness	DeepSeek	1418	88.8
13	MiniMax H3	MiniMax	1412	88.2
14	Kimi K3	Moonshot AI	1408	87.5
15	Muse Wonder Full	Meta	1405	86.8
16	Claude Haiku 5.5	Anthropic	1402	87.9
17	Muse Wonder	Meta	1394	85.2
18	GLM-5.3	Zhipu AI	1391	85.5
19	DeepSeek V4.5	DeepSeek	1385	84.5
20	Qwen3.8 Next	Alibaba	1382	83.6
21	Grok-4 Fast Reasoning	xAI	1375	83.0
22	GLM-5.3 Flash	Zhipu AI	1372	84.1
23	Gemini 3.8 Flash	Google	1368	82.7
24	Mistral Large 3.5	Mistral AI	1365	82.3
25	Gemini 3 Pro	Google	1362	81.0
26	Gemini 3.1 Pro	Google	1358	80.2
27	Qwen3.8 27B	Alibaba	1352	79.8
28	MiMo-V2.6-Pro	Xiaomi	1345	78.0
29	GPT-5.6 Luna	OpenAI	1341	78.4
30	MiniMax H3 Flash	MiniMax	1338	78.8
31	Kimi K3 Flash	Moonshot AI	1325	76.0
32	Qwen3.8 Flash	Alibaba	1312	74.5
33	Gemma 4 27B	Google	1301	72.0
34	Strata 2	Strata Labs	1288	70.5
35	Jev 1	Jev Labs	1271	68.9
36	Clef 2	Clef AI	1259	67.2
37	Mercury 2.5	Inception Labs	1244	62.0
38	Celeris-1	Celeris	1208	58.2
39	Laya 1	Laya Research	1196	55.4
40	Pixel Canary	Google	1141	48.0

Arena rankings FAQ

What is AI arena Elo?

Arena Elo is a rating built from blind human votes between two anonymous AI models. Every win moves a model up and every loss moves it down, so the score reflects collective human preference rather than any single benchmark. Claude Opus 5.5 currently leads at 1489 Elo.

Why do arena rankings differ from benchmark rankings?

Benchmarks measure fixed test sets, while arena votes measure felt quality — formatting, tone, and usefulness of the whole answer. Models with polished long-form output often out-vote models that edge them on paper. The AMR aggregate blends both signals, which is why its order sits between the two.