Chatbot Arena

This leaderboard is based on the following benchmarks.

• Chatbot Arena - a crowdsourced, randomized battle platform for large language models (LLMs). We use 2.8M+ user votes to compute Elo ratings.
• MMLU - a test to measure a model’s multitask accuracy on 57 tasks.
• Arena-Hard-Auto - an automatic evaluation tool for instruction-tuned LLMs.

| Vote | Blog | GitHub | Paper | Dataset | Twitter | Discord |

Best Open LM

Model	Arena Elo	MMLU	License
DeepSeek-R1	1360	90.8	MIT
Gemma-3-27B-it	1340		Gemma
DeepSeek-V3	1318	88.5	DeepSeek
QwQ-32B	1315		Apache 2.0
Llama-3.3-Nemotron-Super-49B-v1	1296	86	Nvidia

Full Leaderboard

Model	Arena Elo	Coding	Vision	Arena Hard	MMLU	Votes	Organization	License
🥇 Gemini-2.5-Pro-Exp-03-25	1443	1427	1327			2540	Google	Proprietary
🥇 Grok-3-Preview-02-24	1404	1414			92.7	10398	xAI	Proprietary
🥇 GPT-4.5-Preview	1398	1405	1269			10615	OpenAI	Proprietary
🥈 Gemini-2.0-Flash-Thinking-Exp-01-21	1381	1364	1276			22659	Google	Proprietary
🥈 Gemini-2.0-Pro-Exp-02-05	1380	1380	1241			20293	Google	Proprietary
🥈 ChatGPT-4o-latest (2025-01-29)	1374	1366	1280			22517	OpenAI	Proprietary
🥈 DeepSeek-R1	1360	1368			90.8	12772	DeepSeek	MIT
🥈 Gemini-2.0-Flash-Exp	1355	1353	1257			22520	Google	Proprietary
🥈 o1-2024-12-17	1351	1359	1230	90.4	91.8	25044	OpenAI	Proprietary
🥈 Qwen2.5-Max	1340	1345				17124	Alibaba	Proprietary
🥈 Gemma-3-27B-it	1340	1307				6974	Google	Gemma
🥈 o3-mini-high	1326	1365				14274	OpenAI	Proprietary
🥉 DeepSeek-V3	1318	1320			88.5	22845	DeepSeek	DeepSeek
🥉 QwQ-32B	1315	1327				4050	Alibaba	Apache 2.0
🥉 Command A (03-2025)	1311	1325				3415	Cohere	CC-BY-NC-4.0
🥉 Qwen-Plus-0125	1310	1320				6059	Alibaba	Proprietary
🥉 Gemini-2.0-Flash-Lite	1310	1318	1149			18090	Google	Proprietary
🥉 GLM-4-Plus-0111	1310	1290				6034	Zhipu	Proprietary
🥉 o1-mini	1304	1353		92		54975	OpenAI	Proprietary
🥉 o3-mini	1304	1349				20765	OpenAI	Proprietary
🥉 Claude 3.7 Sonnet (thinking-32k)	1304	1335				4917	Anthropic	Proprietary
🥉 Step-2-16K-Exp	1304	1295				5131	StepFun	Proprietary
🥉 Hunyuan-TurboS-20250226	1303	1328				2460	Tencent	Proprietary
🥉 Gemini-1.5-Pro-002	1302	1291	1222			58693	Google	Proprietary
🥉 Claude 3.7 Sonnet	1296	1338	1220			10197	Anthropic	Proprietary
🥉 Hunyuan-Turbo-0110	1296	1315				2512	Tencent	Proprietary
🥉 Llama-3.3-Nemotron-Super-49B-v1	1296	1302		88.3		2381	Nvidia	Nvidia
🥉 Grok-2-08-13	1288	1282				67107	xAI	Proprietary
🥉 Yi-Lightning	1287	1302		81.5		28972	01 AI	Proprietary
🥉 GPT-4o-2024-05-13	1285	1293	1206	79.21	88.7	117760	OpenAI	Proprietary
🥉 Claude 3.5 Sonnet (20241022)	1283	1326	1183	85.2	88.7	62603	Anthropic	Proprietary
Deepseek-v2.5-1210	1279	1297				7249	DeepSeek	DeepSeek
Athene-v2-Chat-72B	1275	1300		85		26110	NexusFlow	NexusFlow
Hunyuan-Large-2025-02-10	1272	1293				3859	Tencent	Proprietary
GPT-4o-mini-2024-07-18	1272	1283	1124	74.94	82	69408	OpenAI	Proprietary
Gemini-1.5-Flash-002	1271	1254	1205			37012	Google	Proprietary
Llama-3.1-405B-Instruct-bf16	1268	1279			88.6	39742	Meta	Llama 3.1
Llama-3.1-Nemotron-70B-Instruct	1268	1271		84.9		7578	Nvidia	Llama 3.1
Llama-3.1-405B-Instruct-fp8	1267	1276		69.3	88.6	63067	Meta	Llama 3.1
Grok-2-Mini-08-13	1266	1262				55449	xAI	Proprietary
Yi-Lightning-lite	1264	1267				17071	01 AI	Proprietary
Hunyuan-Standard-2025-02-10	1259	1268				4021	Tencent	Proprietary
Qwen2.5-72B-Instruct	1257	1283		78		41552	Alibaba	Qwen
Llama-3.3-70B-Instruct	1257	1259				34033	Meta	Llama-3.3
GPT-4-Turbo-2024-04-09	1256	1263	1151	82.63		102184	OpenAI	Proprietary
Mistral-Large-2407	1251	1269		70.42		48233	Mistral	Mistral Research
GPT-4-1106-preview	1250	1253				103764	OpenAI	Proprietary
Athene-70B	1250	1253		77.6		20597	NexusFlow	CC-BY-NC-4.0
Mistral-Large-2411	1248	1265				26946	Mistral	MRL
Llama-3.1-70B-Instruct	1248	1251		55.73	86	58683	Meta	Llama 3.1
Claude 3 Opus	1247	1250	1076	60.36	86.8	202710	Anthropic	Proprietary
Amazon Nova Pro 1.0	1246	1262	1044			22197	Amazon	Proprietary
GPT-4-0125-preview	1245	1243		77.96		97094	OpenAI	Proprietary
Llama-3.1-Tulu-3-70B	1244	1233				3015	Ai2	Llama 3.1
Yi-Large-preview	1240	1245		71.48		51652	01 AI	Proprietary
Claude 3.5 Haiku (20241022)	1237	1264				29154	Anthropic	Propretary
Reka-Core-20240904	1235	1222				7941	Reka AI	Proprietary
Reka-Core-20240722	1231	1208				13290	Reka AI	Proprietary
Qwen-Plus-0828	1227	1245				14623	Alibaba	Proprietary
Gemini-1.5-Flash-001	1227	1232	1072	49.61	78.9	65683	Google	Proprietary
Jamba-1.5-Large	1221	1227			81.2	9129	AI21 Labs	Jamba Open
Deepseek-v2-API-0628	1220	1242				19514	DeepSeek AI	DeepSeek
Gemma-2-27B-it	1220	1209		57.51		79537	Google	Gemma license
Qwen2.5-Coder-32B-Instruct	1217	1261				5730	Alibaba	Apache 2.0
Amazon Nova Lite 1.0	1217	1236	1061			20235	Amazon	Proprietary
Gemma-2-9B-it-SimPO	1216	1196				10554	Princeton	MIT
Command R+ (08-2024)	1215	1181				10540	Cohere	CC-BY-NC-4.0
Deepseek-Coder-v2-0724	1214	1266		62.3		11733	DeepSeek	Proprietary
Mistral-Small-24B-Instruct-2501	1214	1227				12426	Mistral	Apache 2.0
Yi-Large	1213	1220		63.7		16631	01 AI	Proprietary
Gemini-1.5-Flash-8B-001	1212	1208	1106			37695	Google	Proprietary
Llama-3.1-Nemotron-51B-Instruct	1211	1211				3886	Nvidia	Llama 3.1
Nemotron-4-340B-Instruct	1209	1198				20608	Nvidia	NVIDIA Open Model
Aya-Expanse-32B	1209	1193				28759	Cohere	CC-BY-NC-4.0
Gemini App (2024-01-24)	1208	1171				11839	Google	Proprietary
Llama-3-70B-Instruct	1207	1199		46.57	82	163682	Meta	Llama 3
GLM-4-0520	1206	1216		63.84		10218	Zhipu AI	Proprietary
Reka-Flash-20240904	1205	1191				8136	Reka AI	Proprietary
Gemini-1.5-Flash-8B-Exp-0827	1205	1189	1112			25357	Google	Proprietary
Phi-4	1204	1223				21497	Microsoft	MIT
Claude 3 Sonnet	1201	1213	1048	46.8	79	113083	Anthropic	Proprietary
Reka-Flash-20240722	1201	1187				13731	Reka AI	Proprietary
Reka-Core-20240501	1199	1190	1015		83.2	62586	Reka AI	Proprietary
Amazon Nova Micro 1.0	1198	1211				20253	Amazon	Proprietary
Gemma-2-9B-it	1192	1173				57201	Google	Gemma license
Command R+ (04-2024)	1190	1164		33.07		80880	Cohere	CC-BY-NC-4.0
Hunyuan-Standard-256K	1189	1227				2902	Tencent	Proprietary
Qwen2-72B-Instruct	1187	1187		46.86	84.2	38888	Alibaba	Qianwen LICENSE
GPT-4-0314	1186	1195		50	86.4	55981	OpenAI	Proprietary
Llama-3.1-Tulu-3-8B	1185	1179				3073	Ai2	Llama 3.1
GLM-4-0116	1183	1191		55.72		7580	Zhipu AI	Proprietary
Qwen-Max-0428	1183	1189				25693	Alibaba	Proprietary
Ministral-8B-2410	1182	1201				5115	Mistral	MRL
Aya-Expanse-8B	1180	1165				10390	Cohere	CC-BY-NC-4.0
Claude 3 Haiku	1179	1189	1000	41.47	75.2	122349	Anthropic	Proprietary
Command R (08-2024)	1179	1161				10849	Cohere	CC-BY-NC-4.0
DeepSeek-Coder-V2-Instruct	1178	1239				15753	DeepSeek AI	DeepSeek License
Llama-3.1-8B-Instruct	1176	1186		21.34	73	52607	Meta	Llama 3.1
Jamba-1.5-Mini	1176	1181			69.7	9272	AI21 Labs	Jamba Open
Reka-Flash-Preview-20240611	1165	1155	1024			20430	Reka AI	Proprietary
GPT-4-0613	1163	1167		37.9		91641	OpenAI	Proprietary
Qwen1.5-110B-Chat	1161	1174			80.4	27440	Alibaba	Qianwen LICENSE
Mistral-Large-2402	1157	1169		37.71	81.2	64931	Mistral	Proprietary
Yi-1.5-34B-Chat	1157	1162			76.8	25138	01 AI	Apache-2.0
Reka-Flash-21B-online	1156	1147				16022	Reka AI	Proprietary
QwQ-32B-Preview	1153	1147				3409	Alibaba	Apache 2.0
Llama-3-8B-Instruct	1152	1146		20.56	68.4	109093	Meta	Llama 3
InternLM2.5-20B-chat	1149	1158				10604	InternLM	Other
Claude-1	1149	1135			77	21151	Anthropic	Proprietary
Command R (04-2024)	1149	1123		17.02		56386	Cohere	CC-BY-NC-4.0
Mistral Medium	1148	1152		31.9	75.3	35561	Mistral	Proprietary
Qwen1.5-72B-Chat	1147	1160		36.12	77.5	40666	Alibaba	Qianwen LICENSE
Mixtral-8x22b-Instruct-v0.1	1147	1152		36.36	77.8	53769	Mistral	Apache 2.0
Reka-Flash-21B	1147	1141			73.5	25807	Reka AI	Proprietary
Gemma-2-2b-it	1144	1107			51.3	48921	Google	Gemma license
Granite-3.1-8B-Instruct	1142	1172				3299	IBM	Apache 2.0
Claude-2.0	1132	1135		23.99	78.5	12758	Anthropic	Proprietary
Gemini-1.0-Pro-001	1131	1103			71.8	18809	Google	Proprietary
Zephyr-ORPO-141b-A35b-v0.1	1127	1124				4862	HuggingFace	Apache 2.0
Qwen1.5-32B-Chat	1125	1149			73.4	22769	Alibaba	Qianwen LICENSE
Mistral-Next	1124	1132		27.37		12377	Mistral	Proprietary
Phi-3-Medium-4k-Instruct	1123	1125		33.37	78	26107	Microsoft	MIT
Granite-3.1-2B-Instruct	1119	1147				3384	IBM	Apache 2.0
Starling-LM-7B-beta	1119	1129		23.01		16676	Nexusflow	Apache-2.0
Claude-2.1	1118	1132		22.77		37697	Anthropic	Proprietary
GPT-3.5-Turbo-0613	1117	1135		24.82		38958	OpenAI	Proprietary
Mixtral-8x7B-Instruct-v0.1	1114	1114		23.4	70.6	76141	Mistral	Apache 2.0
Claude-Instant-1	1111	1108			73.4	20625	Anthropic	Proprietary
Yi-34B-Chat	1111	1106		23.15	73.5	15925	01 AI	Yi License
Gemini Pro	1111	1092		17.8	71.8	6559	Google	Proprietary
Qwen1.5-14B-Chat	1109	1126			67.6	18686	Alibaba	Qianwen LICENSE
GPT-3.5-Turbo-0314	1107	1115		18.05	70	5639	OpenAI	Proprietary
GPT-3.5-Turbo-0125	1106	1124		23.34		68884	OpenAI	Proprietary
WizardLM-70B-v1.0	1106	1071			63.7	8382	Microsoft	Llama 2
DBRX-Instruct-Preview	1103	1118		24.63	73.7	33739	Databricks	DBRX LICENSE
Llama-3.2-3B-Instruct	1103	1080				8394	Meta	Llama 3.2
Phi-3-Small-8k-Instruct	1102	1107		29.77	75.7	18476	Microsoft	MIT
Tulu-2-DPO-70B	1099	1093		14.99		6659	AllenAI/UW	AI2 ImpACT Low-risk
Granite-3.0-8B-Instruct	1093	1097				7002	IBM	Apache 2.0
Llama-2-70B-chat	1093	1072		11.55	63	39617	Meta	Llama 2
OpenChat-3.5-0106	1091	1102			65.8	12990	OpenChat	Apache-2.0
Snowflake Arctic Instruct	1090	1077		17.61	67.3	34175	Snowflake	Apache 2.0
Vicuna-33B	1090	1067		8.63	59.2	22941	LMSYS	Non-commercial
Starling-LM-7B-alpha	1088	1080		12.8	63.9	10420	UC Berkeley	CC-BY-NC-4.0
Gemma-1.1-7B-it	1084	1084		12.09	64.3	25066	Google	Gemma license
Nous-Hermes-2-Mixtral-8x7B-DPO	1084	1079				3837	NousResearch	Apache-2.0
NV-Llama2-70B-SteerLM-Chat	1081	1022			68.5	3637	Nvidia	Llama 2
pplx-70B-online	1078	1028				6897	Perplexity AI	Proprietary
DeepSeek-LLM-67B-Chat	1077	1079			71.3	4988	DeepSeek AI	DeepSeek License
OpenChat-3.5	1076	1054			64.3	8110	OpenChat	Apache-2.0
Granite-3.0-2B-Instruct	1074	1088				7190	IBM	Apache 2.0
OpenHermes-2.5-Mistral-7B	1074	1057				5089	NousResearch	Apache-2.0
Mistral-7B-Instruct-v0.2	1072	1074		12.57		20066	Mistral	Apache-2.0
Phi-3-Mini-4K-Instruct-June-24	1071	1082			70.9	12811	Microsoft	MIT
Qwen1.5-7B-Chat	1070	1089			61	4872	Alibaba	Qianwen LICENSE
GPT-3.5-Turbo-1106	1068	1095		18.87		17037	OpenAI	Proprietary
Phi-3-Mini-4k-Instruct	1066	1086			68.8	21099	Microsoft	MIT
Llama-2-13b-chat	1063	1051			53.6	19724	Meta	Llama 2
Dolphin-2.2.1-Mistral-7B	1063	1025				1713	Cognitive Computations	Apache-2.0
SOLAR-10.7B-Instruct-v1.0	1062	1047			66.2	4289	Upstage AI	CC-BY-NC-4.0
WizardLM-13b-v1.2	1059	1025			52.7	7175	Microsoft	Llama 2
Llama-3.2-1B-Instruct	1054	1046				8524	Meta	Llama 3.2
Step-1o-Vision-32k (highres)			1186				StepFun	Proprietary
Qwen2.5-VL-72B-Instruct			1168				Alibaba	Qwen
Pixtral-Large-2411			1154				Mistral	MRL
Qwen-VL-Max-1119			1128				Alibaba	Proprietary
Step-1V-32K			1111				StepFun	Proprietary
Qwen2-VL-72b-Instruct			1110				Alibaba	Qwen
Molmo-72B-0924			1075				AI2	Apache 2.0
Pixtral-12B-2409			1072				Mistral	Apache 2.0
Llama-3.2-90B-Vision-Instruct			1069				Meta	Llama 3.2
InternVL2-26B			1067				OpenGVLab	MIT
Hunyuan-Standard-Vision-2024-12-31			1064				Tencent	Proprietary
Qwen2-VL-7B-Instruct			1054				Aliaba	Apache 2.0
Yi-Vision			1045				01 AI	Proprietary
Llama-3.2-11B-Vision-Instruct			1032				Meta	Llama 3.2

If you want to see more models, please help us add them.

💻 Code: The Arena Elo ratings are computed by this notebook. The MT-bench scores (single-answer grading on a scale of 10) are computed by fastchat.llm_judge. The MMLU scores are computed by InstructEval. Higher values are better for all benchmarks. Empty cells mean not available. The latest and detailed leaderboard is here.

More Statistics for Chatbot Arena

🔗 Arena Statistics

Transition from online Elo rating system to Bradley-Terry model

We adopted the Elo rating system for ranking models since the launch of the Arena. It has been useful to transform pairwise human preference to Elo ratings that serve as a predictor of winrate between models. Specifically, if player A has a rating of R_A and player B a rating of R_B, the probability of player A winning is

ELO rating has been used to rank chess players by the international community for over 60 years. Standard Elo rating systems assume a player’s performance changes overtime. So an online algorithm is needed to capture such dynamics, meaning recent games should weigh more than older games. Specifically, after each game, a player’s rating is updated according to the difference between predicted outcome and actual outcome.

This algorithm has two distinct features:

1. It can be computed asynchronously by players around the world.
2. It allows for players performance to change dynamically – it does not assume a fixed unknown value for the players rating.

This ability to adapt is determined by the parameter K which controls the magnitude of rating changes that can affect the overall result. A larger K essentially put more weight on the recent games, which may make sense for new players whose performance improves quickly. However as players become more senior and their performance “converges” then a smaller value of K is more appropriate. As a result, USCF adopted K based on the number of games and tournaments completed by the player (reference). That is, the Elo rating of a senior player changes slower than a new player.

When we launched the Arena, we noticed considerable variability in the ratings using the classic online algorithm. We tried to tune the K to be sufficiently stable while also allowing new models to move up quickly in the leaderboard. We ultimately decided to adopt a bootstrap-like technique to shuffle the data and sample Elo scores from 1000 permutations of the online plays. You can find the details in this notebook. This provided consistent stable scores and allowed us to incorporate new models quickly. This is also observed in a recent work by Cohere. However, we used the same samples to estimate confidence intervals which were therefore too wide (effectively CI’s for the original online Elo estimates).

In the context of LLM ranking, there are two important differences from the classic Elo chess ranking system. First, we have access to the entire history of all games for all models and so we don’t need a decentralized algorithm. Second, most models are static (we have access to the weights) and so we don’t expect their performance to change. However, it is worth noting that the hosted proprietary models may not be static and their behavior can change without notice. We try our best to pin specific model API versions if possible.

To improve the quality of our rankings and their confidence estimates, we are adopting another widely used rating system called the Bradley–Terry (BT) model. This model actually is the maximum likelihood (MLE) estimate of the underlying Elo model assuming a fixed but unknown pairwise win-rate. Similar to Elo rating, BT model is also based on pairwise comparison to derive ratings of players to estimate win rate between each other. The core difference between BT model vs the online Elo system is the assumption that player’s performance does not change (i.e., game order does not matter) and the computation takes place in a centralized fashion.