Skip to content
IFAI

Model rankings

505 models · 3 boards · 2 sources

Artificial AnalysisIntelligence IndexArenaText arena

Text arena · Anonymous head-to-head votes · Published August 21, 2025

Most preferred
gemini-2.5-proGoogle1,457Score
Best open model
Best from China
Most voted
claude-3-opus-20240229Anthropic202,641Votes
Top three by Score
1gemini-2.5-proGoogle1,457
2gpt-5-highOpenAI1,455
Top three open models
1qwen3-235b-a22b-instruct-2507Alibaba · #10 overall1,422
2kimi-k2-0711-previewMoonshot AI · #11 overall1,421
3deepseek-r1-0528DeepSeek · #13 overall1,417
This week's changes
Mistral Medium 3.1Intelligence · #58New
Gemma 3 4B InstructIntelligence · #132Up 13
Gemini 2.5 Flash-Lite (Reasoning)Intelligence · #32Up 7
GPT-4.1 miniIntelligence · #38Up 7
GPT-4.1 nanoIntelligence · #76Up 3
RankChangeModelScore95% intervalVotesPrice $/M
1—qwen3-235b-a22b-instruct-2507OpenAlibaba1,422±69,744—
2—kimi-k2-0711-previewOpenMoonshot AI1,421±615,926—
3—deepseek-r1-0528OpenDeepSeek · 1 more variant1,417±621,205—
4—glm-4.5OpenZ.ai1,413±78,707—
5—qwen3-235b-a22b-thinking-2507OpenAlibaba1,403±76,996—
6—qwen3-235b-a22b-no-thinkingOpenAlibaba1,397±527,765—
7—deepseek-v3-0324OpenDeepSeek · 1 more variant1,393±530,936—
8—qwen3-coder-480b-a35b-instructOpenAlibaba1,382±69,603—
9—qwen3-30b-a3b-instruct-2507OpenAlibaba1,380±76,951—
10—glm-4.5-airOpenZ.ai1,378±78,075—
11—qwen3-235b-a22bOpenAlibaba1,375±523,704—
12—minimax-m1OpenMiniMax1,370±520,987—
13—gemma-3-27b-itOpenGoogle1,364±435,379—
14—gpt-oss-120bOpenOpenAI1,354±86,099$0.26
15▲ 4step-3OpenStepFun1,354±112,973—
16▼ 1mistral-small-2506OpenMistral1,353±612,660—
17▼ 1command-a-03-2025OpenCohere1,347±434,458—
18▼ 1qwen3-32bOpenAlibaba1,346±94,074—
19▼ 1llama-3.1-nemotron-ultra-253b-v1OpenNVIDIA1,345±112,656—
20—nvidia-llama-3.3-nemotron-super-49b-v1.5OpenNVIDIA1,342±93,917—
21—gemma-3-12b-itOpenGoogle1,341±93,976—
22▲ 10glm-4.5vOpenZ.ai1,339±18992—
23▼ 1qwq-32bOpenAlibaba · 1 more variant1,337±524,118—
24▼ 1llama-3.1-405b-instruct-bf16OpenMeta1,335±443,788—
25▼ 1llama-3.1-405b-instruct-fp8OpenMeta1,333±363,038—
26▼ 1gpt-oss-20bOpenOpenAI1,332±85,721—
27▼ 1llama-4-maverick-17b-128e-instructOpenMeta1,328±528,106—
28▼ 1qwen3-30b-a3bOpenAlibaba1,327±523,737—
29▼ 1llama-3.3-nemotron-49b-super-v1OpenNVIDIA1,325±122,371—
30▼ 1deepseek-v2.5-1210OpenDeepSeek · 1 more variant1,322±87,243—
31—llama-3.3-70b-instructOpenMeta1,320±354,132—
32▼ 2llama-4-scout-17b-16e-instructOpenMeta1,319±617,813—
33—qwen-max-0919OpenAlibaba1,317±517,432—
34—gemma-3n-e4b-itOpenGoogle1,316±616,502—
35—athene-v2-chatOpenNexusFlow1,314±426,074—
36—mistral-large-2407OpenMistral · 1 more variant1,313±448,217—
37—athene-70b-0725OpenNexusFlow1,305±520,580—
38—gemma-3-4b-itOpenGoogle1,303±94,321—
39—qwen2.5-72b-instructOpenAlibaba1,302±441,519—
40—llama-3.1-nemotron-70b-instructOpenNVIDIA1,298±77,577—
41—mistral-small-3.1-24b-instruct-2503OpenMistral1,296±617,533—
42—llama-3.1-70b-instructOpenMeta1,294±358,637—
43—llama-3.1-tulu-3-70bOpenAi21,289±103,010—
44—jamba-1.5-largeOpenAI21 Labs1,289±79,125—
45—llama-3.1-nemotron-51b-instructOpenNVIDIA1,286±103,889—
46—gemma-2-27b-itOpenGoogle1,286±379,538—
47—gemma-2-9b-it-simpoOpenPrinceton1,280±710,548—
48—nemotron-4-340b-instructOpenNVIDIA1,279±520,608—
49—command-r-plus-08-2024OpenCohere1,279±610,535—
50—llama-3-70b-instructOpenMeta1,276±3163,629—

Intelligence over time

Each maker's best model at the time on the Artificial Analysis Intelligence Index, by the month it was released.

Gained most · Anthropic +57.2Highest now · Anthropic 64.6Gap between highest and lowest 18.1 ptsTime range Nov 2025 – Nov 2026
64.6Anthropic60.0Google56.9OpenAI50.1Meta46.5Kimi

Arena data comes from the leaderboard dataset it publishes on Hugging Face (CC BY 4.0). Only each board's overall leaderboard is used; scores and ranks are unchanged. dataset · CC BY 4.0

Artificial Analysis data comes from its API (artificialanalysis.ai). Scores are unchanged; ranks follow the scores. artificialanalysis.ai

Open or closed follows the license Arena gives; prices and Artificial Analysis intelligence show only where model names match, otherwise “—”.