Arene documents

De Wikiprompt, l’encyclopedie libre des prompts

Classement Elo communautaire, agrege chaque jour depuis les duels LMArena. L’arene generale d’abord, puis le detail par categorie.

Derniere mise a jour: 2026-08-24 · 38 modeles

Rang	Modele	Organisation	Score Elo	IC 95%	Votes	Licence
1	claude-opus-5-high	anthropic	1,520	1,505 / 1,535	1,663	Proprietary
2	claude-opus-4-6	anthropic	1,510	1,504 / 1,516	37,271	Proprietary
3	claude-opus-4-6-thinking	anthropic	1,506	1,499 / 1,513	24,973	Proprietary
4	claude-fable-5	anthropic	1,504	1,495 / 1,513	5,792	Proprietary
5	claude-opus-4-7	anthropic	1,498	1,491 / 1,505	18,990	Proprietary
6	claude-opus-4-7-thinking	anthropic	1,497	1,490 / 1,504	18,793	Proprietary
7	gpt-5.5-high	openai	1,485	1,478 / 1,492	16,816	Proprietary
8	claude-sonnet-4-6	anthropic	1,483	1,476 / 1,489	54,427	Proprietary
9	gpt-5.5	openai	1,480	1,473 / 1,487	17,286	Proprietary
10	gpt-5.6-terra-xhigh	openai	1,479	1,466 / 1,492	1,969	Proprietary
11	gpt-5.6-sol-xhigh	openai	1,479	1,462 / 1,495	1,152	Proprietary
12	claude-opus-4-8-thinking	anthropic	1,475	1,466 / 1,483	8,388	Proprietary
13	muse-spark-1.1	meta	1,472	1,459 / 1,485	2,054	Proprietary
14	claude-sonnet-5-high	anthropic	1,470	1,460 / 1,481	3,601	Proprietary
15	gpt-5.4	openai	1,470	1,464 / 1,477	29,809	Proprietary
16	claude-opus-4-8	anthropic	1,469	1,461 / 1,477	8,193	Proprietary
17	gemini-3.5-flash-medium	google	1,465	1,454 / 1,475	3,689	Proprietary
18	gpt-5.6-luna-xhigh	openai	1,462	1,449 / 1,476	1,888	Proprietary
19	claude-opus-4-5-20251101	anthropic	1,462	1,451 / 1,472	7,965	Proprietary
20	grok-4.5	xai	1,454	1,442 / 1,466	2,435	Proprietary
21	kimi-k2.6	moonshot	1,451	1,443 / 1,459	11,181	Modified MIT
22	claude-sonnet-4-5-20250929	anthropic	1,446	1,440 / 1,453	28,634	Proprietary
23	gemini-3.1-pro-preview	google	1,445	1,439 / 1,451	45,162	Proprietary
24	muse-spark	meta	1,443	1,425 / 1,462	1,078	Proprietary
25	qwen3.7-plus	alibaba	1,440	1,429 / 1,450	3,088	Proprietary
26	minimax-m3	minimax	1,435	1,426 / 1,443	6,263	MiniMax Community License
27	gemini-3-pro	google	1,433	1,424 / 1,442	10,739	Proprietary
28	kimi-k2.5-thinking	moonshot	1,429	1,423 / 1,436	19,891	Modified MIT
29	gemma-4-31b	google	1,424	1,416 / 1,432	10,577	Apache 2.0
30	claude-haiku-4-5-20251001	anthropic	1,423	1,417 / 1,429	30,937	Proprietary
31	gemini-2.5-pro	google	1,421	1,415 / 1,427	24,963	Proprietary
32	glm-5v-turbo	zai	1,417	1,408 / 1,427	4,894	Proprietary
33	grok-4.20-beta-0309-reasoning	xai	1,414	1,407 / 1,421	18,666	Proprietary
34	gemini-3-flash	google	1,413	1,404 / 1,423	7,173	Proprietary
35	gpt-5.2-high	openai	1,405	1,395 / 1,415	7,073	Proprietary
36	gpt-5.5-instant	openai	1,402	1,394 / 1,410	8,442	Proprietary
37	gpt-5.1	openai	1,401	1,391 / 1,410	8,220	Proprietary
38	gpt-5.2	openai	1,400	1,394 / 1,406	28,068	Proprietary

Voir aussi

Sources et methodologie

Les classements sont agreges depuis des evaluations communautaires ouvertes. Les scores Elo, intervalles de confiance et votes proviennent de LMArena (duels de modeles votes par la communaute, publies en dataset ouvert). Les prix et metadonnees de contexte proviennent de l’API publique OpenRouter. Wikiprompt prend un instantane quotidien et preserve l’historique ; nous ne menons aucune evaluation proprietaire.