Chatbot Arena

This leaderboard is based on the following benchmarks.

• Chatbot Arena - a crowdsourced, randomized battle platform for large language models (LLMs). We use 2.5M+ user votes to compute Elo ratings.
• MMLU (5-shot) - 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-V3	1320	88.5	DeepSeek
Qwen2.5-72B-Instruct	1257	86.8	Qwen
Llama-3.3-70B-Instruct	1256	86	Llama 3.3

Full Leaderboard

Model	Arena Elo	Coding Elo	Arena Hard	MMLU	Votes	Organization	License
🥇 Gemini-2.0-Flash-Thinking-Exp-01-21	1380	1357			5572	Google	Proprietary
🥇 Gemini-Exp-1206	1374	1369			21004	Google	Proprietary
🥇 ChatGPT-4o-latest (2024-11-20)	1365	1352			34209	OpenAI	Proprietary
🥇 Gemini-2.0-Flash-Exp	1356	1351			19823	Google	Proprietary
🥇 o1-2024-12-17	1351	1362	90.4		8124	OpenAI	Proprietary
🥈 DeepSeek-V3	1320	1317			11893	DeepSeek	DeepSeek
🥈 o1-mini	1306	1356	92		48847	OpenAI	Proprietary
🥈 Step-2-16K-Exp	1306	1301			4106	StepFun	Proprietary
🥈 Gemini-1.5-Pro-002	1303	1289			45406	Google	Proprietary
🥈 Grok-2-08-13	1288	1282			66412	xAI	Proprietary
🥈 Yi-Lightning	1287	1302	81.5		28955	01 AI	Proprietary
🥈 Claude 3.5 Sonnet (20241022)	1283	1324	85.2	88.7	47647	Anthropic	Proprietary
🥈 Qwen2.5-plus-1127	1282	1302			7872	Alibaba	Proprietary
🥉 Deepseek-v2.5-1210	1279	1296			7259	DeepSeek	DeepSeek
🥉 Athene-v2-Chat-72B	1277	1298	85		21222	NexusFlow	NexusFlow
🥉 GLM-4-Plus	1274	1283			27775	Zhipu AI	Proprietary
🥉 GPT-4o-mini-2024-07-18	1273	1284	74.94	82	60623	OpenAI	Proprietary
🥉 Gemini-1.5-Flash-002	1271	1253			34631	Google	Proprietary
🥉 Llama-3.1-Nemotron-70B-Instruct	1269	1271	84.9		7595	Nvidia	Llama 3.1
🥉 Llama-3.1-405B-Instruct	1268	1279		88.6	21520	Meta	Llama 3.1
🥉 Grok-2-Mini-08-13	1266	1262			54988	xAI	Proprietary
🥉 Yi-Lightning-lite	1264	1267			17062	01 AI	Proprietary
🥉 Qwen2.5-72B-Instruct	1257	1282	78		40070	Alibaba	Qwen
🥉 GPT-4-Turbo-2024-04-09	1256	1263	82.63		102127	OpenAI	Proprietary
🥉 Llama-3.3-70B-Instruct	1256	1261			15711	Meta	Llama-3.3
Mistral-Large-2407	1252	1269	70.42		48209	Mistral	Mistral Research
Athene-70B	1250	1254	77.6		20620	NexusFlow	CC-BY-NC-4.0
GPT-4-1106-preview	1250	1253			103733	OpenAI	Proprietary
Llama-3.1-70B-Instruct	1248	1251	55.73	86	58789	Meta	Llama 3.1
Claude 3 Opus	1247	1250	60.36	86.8	202735	Anthropic	Proprietary
GPT-4-0125-preview	1245	1244	77.96		97076	OpenAI	Proprietary
Amazon Nova Pro 1.0	1244	1257			13149	Amazon	Proprietary
Llama-3.1-Tulu-3-70B	1244	1231			3029	Ai2	Llama 3.1
Mistral-Large-2411	1243	1260			10831	Mistral	MRL
Yi-Large-preview	1240	1245	71.48		51657	01 AI	Proprietary
Claude 3.5 Haiku (20241022)	1238	1261			10995	Anthropic	Propretary
Reka-Core-20240904	1235	1222			7940	Reka AI	Proprietary
Reka-Core-20240722	1230	1208			13295	Reka AI	Proprietary
Qwen-Plus-0828	1227	1245			14631	Alibaba	Proprietary
Gemini-1.5-Flash-001	1227	1232	49.61	78.9	65648	Google	Proprietary
Jamba-1.5-Large	1221	1227		81.2	9124	AI21 Labs	Jamba Open
Deepseek-v2-API-0628	1220	1242			19516	DeepSeek AI	DeepSeek
Gemma-2-27B-it	1220	1211	57.51		72521	Google	Gemma license
Amazon Nova Lite 1.0	1218	1234			10969	Amazon	Proprietary
Qwen2.5-Coder-32B-Instruct	1217	1260			5731	Alibaba	Apache 2.0
Gemma-2-9B-it-SimPO	1216	1197			10555	Princeton	MIT
Command R+ (08-2024)	1215	1181			10547	Cohere	CC-BY-NC-4.0
Deepseek-Coder-v2-0724	1214	1266	62.3		11727	DeepSeek	Proprietary
Yi-Large	1213	1220	63.7		16634	01 AI	Proprietary
Gemini-1.5-Flash-8B-001	1212	1208			36286	Google	Proprietary
Llama-3.1-Nemotron-51B-Instruct	1211	1211			3897	Nvidia	Llama 3.1
Phi-4	1210	1251			3077	Microsoft	MIT
Nemotron-4-340B-Instruct	1209	1198			20609	Nvidia	NVIDIA Open Model
Aya-Expanse-32B	1209	1192			27096	Cohere	CC-BY-NC-4.0
Gemini App (2024-01-24)	1208	1171			11830	Google	Proprietary
GLM-4-0520	1207	1216	63.84		10214	Zhipu AI	Proprietary
Llama-3-70B-Instruct	1206	1200	46.57	82	163792	Meta	Llama 3
Reka-Flash-20240904	1205	1191			8129	Reka AI	Proprietary
Gemini-1.5-Flash-8B-Exp-0827	1205	1189			25358	Google	Proprietary
Claude 3 Sonnet	1201	1213	46.8	79	113032	Anthropic	Proprietary
Reka-Flash-20240722	1201	1187			13732	Reka AI	Proprietary
Reka-Core-20240501	1200	1190		83.2	62593	Reka AI	Proprietary
Amazon Nova Micro 1.0	1197	1212			11059	Amazon	Proprietary
Gemma-2-9B-it	1191	1173			50349	Google	Gemma license
Command R+ (04-2024)	1190	1164	33.07		80862	Cohere	CC-BY-NC-4.0
Hunyuan-Standard-256K	1189	1227			2902	Tencent	Proprietary
Qwen2-72B-Instruct	1187	1187	46.86	84.2	38887	Alibaba	Qianwen LICENSE
GPT-4-0314	1186	1196	50	86.4	55965	OpenAI	Proprietary
Llama-3.1-Tulu-3-8B	1185	1177			3081	Ai2	Llama 3.1
GLM-4-0116	1183	1191	55.72		7582	Zhipu AI	Proprietary
Qwen-Max-0428	1183	1190			25706	Alibaba	Proprietary
Ministral-8B-2410	1182	1201			5112	Mistral	MRL
Aya-Expanse-8B	1180	1166			8888	Cohere	CC-BY-NC-4.0
Command R (08-2024)	1180	1162			10847	Cohere	CC-BY-NC-4.0
Claude 3 Haiku	1179	1189	41.47	75.2	122294	Anthropic	Proprietary
DeepSeek-Coder-V2-Instruct	1178	1239			15752	DeepSeek AI	DeepSeek License
Llama-3.1-8B-Instruct	1176	1186	21.34	73	52654	Meta	Llama 3.1
Jamba-1.5-Mini	1176	1180		69.7	9269	AI21 Labs	Jamba Open
Reka-Flash-Preview-20240611	1165	1155			20421	Reka AI	Proprietary
GPT-4-0613	1163	1167	37.9		91642	OpenAI	Proprietary
Qwen1.5-110B-Chat	1161	1175		80.4	27462	Alibaba	Qianwen LICENSE
Mistral-Large-2402	1157	1170	37.71	81.2	64920	Mistral	Proprietary
Yi-1.5-34B-Chat	1157	1162		76.8	25128	01 AI	Apache-2.0
Reka-Flash-21B-online	1156	1147			16033	Reka AI	Proprietary
QwQ-32B-Preview	1153	1146			3414	Alibaba	Apache 2.0
Llama-3-8B-Instruct	1152	1146	20.56	68.4	109215	Meta	Llama 3
InternLM2.5-20B-chat	1149	1158			10601	InternLM	Other
Claude-1	1149	1136		77	21162	Anthropic	Proprietary
Command R (04-2024)	1149	1123	17.02		56371	Cohere	CC-BY-NC-4.0
Mixtral-8x22b-Instruct-v0.1	1148	1153	36.36	77.8	53790	Mistral	Apache 2.0
Mistral Medium	1148	1153	31.9	75.3	35557	Mistral	Proprietary
Reka-Flash-21B	1148	1141		73.5	25805	Reka AI	Proprietary
Qwen1.5-72B-Chat	1147	1160	36.12	77.5	40639	Alibaba	Qianwen LICENSE
Gemma-2-2b-it	1142	1107		51.3	41950	Google	Gemma license
Granite-3.1-8B-Instruct	1139	1173			2583	IBM	Apache 2.0
Claude-2.0	1132	1135	23.99	78.5	12757	Anthropic	Proprietary
Gemini-1.0-Pro-001	1131	1103		71.8	18796	Google	Proprietary
Zephyr-ORPO-141b-A35b-v0.1	1127	1124			4859	HuggingFace	Apache 2.0
Qwen1.5-32B-Chat	1125	1149		73.4	22768	Alibaba	Qianwen LICENSE
Mistral-Next	1124	1132	27.37		12376	Mistral	Proprietary
Phi-3-Medium-4k-Instruct	1123	1125	33.37	78	26110	Microsoft	MIT
Starling-LM-7B-beta	1119	1129	23.01		16672	Nexusflow	Apache-2.0
Claude-2.1	1118	1132	22.77		37700	Anthropic	Proprietary
Granite-3.1-2B-Instruct	1117	1148			2604	IBM	Apache 2.0
GPT-3.5-Turbo-0613	1117	1135	24.82		38946	OpenAI	Proprietary
Mixtral-8x7B-Instruct-v0.1	1114	1114	23.4	70.6	76140	Mistral	Apache 2.0
Claude-Instant-1	1111	1109		73.4	20620	Anthropic	Proprietary
Yi-34B-Chat	1111	1106	23.15	73.5	15920	01 AI	Yi License
Gemini Pro	1110	1092	17.8	71.8	6559	Google	Proprietary
Qwen1.5-14B-Chat	1109	1126		67.6	18671	Alibaba	Qianwen LICENSE
GPT-3.5-Turbo-0125	1106	1124	23.34		68863	OpenAI	Proprietary
GPT-3.5-Turbo-0314	1106	1115	18.05	70	5645	OpenAI	Proprietary
WizardLM-70B-v1.0	1106	1071		63.7	8380	Microsoft	Llama 2
DBRX-Instruct-Preview	1103	1118	24.63	73.7	33731	Databricks	DBRX LICENSE
Llama-3.2-3B-Instruct	1103	1080			8409	Meta	Llama 3.2
Phi-3-Small-8k-Instruct	1102	1107	29.77	75.7	18479	Microsoft	MIT
Tulu-2-DPO-70B	1099	1093	14.99		6663	AllenAI/UW	AI2 ImpACT Low-risk
Granite-3.0-8B-Instruct	1093	1097			7003	IBM	Apache 2.0
Llama-2-70B-chat	1093	1072	11.55	63	39634	Meta	Llama 2
OpenChat-3.5-0106	1091	1102		65.8	12981	OpenChat	Apache-2.0
Vicuna-33B	1091	1067	8.63	59.2	22951	LMSYS	Non-commercial
Snowflake Arctic Instruct	1090	1077	17.61	67.3	34176	Snowflake	Apache 2.0
Starling-LM-7B-alpha	1088	1080	12.8	63.9	10416	UC Berkeley	CC-BY-NC-4.0
Gemma-1.1-7B-it	1084	1084	12.09	64.3	25063	Google	Gemma license
Nous-Hermes-2-Mixtral-8x7B-DPO	1084	1079			3834	NousResearch	Apache-2.0
NV-Llama2-70B-SteerLM-Chat	1081	1023		68.5	3637	Nvidia	Llama 2
pplx-70B-online	1078	1028			6891	Perplexity AI	Proprietary
DeepSeek-LLM-67B-Chat	1077	1079		71.3	4987	DeepSeek AI	DeepSeek License
OpenChat-3.5	1076	1054		64.3	8112	OpenChat	Apache-2.0
OpenHermes-2.5-Mistral-7B	1075	1058			5091	NousResearch	Apache-2.0
Granite-3.0-2B-Instruct	1074	1088			7192	IBM	Apache 2.0
Mistral-7B-Instruct-v0.2	1072	1074	12.57		20056	Mistral	Apache-2.0
Phi-3-Mini-4K-Instruct-June-24	1071	1082		70.9	12820	Microsoft	MIT
Qwen1.5-7B-Chat	1070	1089		61	4869	Alibaba	Qianwen LICENSE
GPT-3.5-Turbo-1106	1068	1095	18.87		17031	OpenAI	Proprietary
Phi-3-Mini-4k-Instruct	1066	1086		68.8	21086	Microsoft	MIT
Llama-2-13b-chat	1063	1051		53.6	19737	Meta	Llama 2
Dolphin-2.2.1-Mistral-7B	1063	1026			1713	Cognitive Computations	Apache-2.0
SOLAR-10.7B-Instruct-v1.0	1062	1047		66.2	4288	Upstage AI	CC-BY-NC-4.0
WizardLM-13b-v1.2	1059	1026		52.7	7178	Microsoft	Llama 2
Llama-3.2-1B-Instruct	1054	1046			8535	Meta	Llama 3.2
Zephyr-7B-beta	1053	1030		61.4	11335	HuggingFace	MIT
SmolLM2-1.7B-Instruct	1047	1043			2371	HuggingFace	Apache 2.0
MPT-30B-chat	1045	1031		50.4	2649	MosaicML	CC-BY-NC-SA-4.0
pplx-7B-online	1044	1015			6336	Perplexity AI	Proprietary
CodeLlama-34B-instruct	1043	1042		53.7	7515	Meta	Llama 2
CodeLlama-70B-instruct	1042	1048			1193	Meta	Llama 2
Zephyr-7B-alpha	1042	1034			1814	HuggingFace	MIT
Vicuna-13B	1042	1032		55.8	19790	LMSYS	Llama 2
Gemma-7B-it	1037	1047		64.3	9176	Google	Gemma license
Phi-3-Mini-128k-Instruct	1037	1029		68.1	21626	Microsoft	MIT
Llama-2-7B-chat	1037	1002		45.8	14550	Meta	Llama 2
Qwen-14B-Chat	1035	1056		66.5	5070	Alibaba	Qianwen LICENSE
falcon-180b-chat	1034	1018		68	1327	TII	Falcon-180B TII License
Guanaco-33B	1032	965		57.6	2999	UW	Non-commercial
Gemma-1.1-2b-it	1021	1036		64.3	11347	Google	Gemma license
StripedHyena-Nous-7B	1018	1000			5273	Together AI	Apache 2.0
OLMo-7B-instruct	1016	1017			6505	Allen AI	Apache-2.0
Mistral-7B-Instruct-v0.1	1008	1008		55.4	9144	Mistral	Apache 2.0
Vicuna-7B	1005	981		49.8	7015	LMSYS	Llama 2
PaLM-Chat-Bison-001	1004	990			8745	Google	Proprietary
Gemma-2B-it	989	1001		42.3	4921	Google	Gemma license
Qwen1.5-4B-Chat	988	990		56.1	7810	Alibaba	Qianwen LICENSE
Koala-13B	964	937		44.7	7035	UC Berkeley	Non-commercial
ChatGLM3-6B	955	953			4765	Tsinghua	Apache-2.0
GPT4All-13B-Snoozy	932	910		43	1786	Nomic AI	Non-commercial
MPT-7B-Chat	928	899		32	4013	MosaicML	CC-BY-NC-SA-4.0
ChatGLM2-6B	924	891		45.5	2707	Tsinghua	Apache-2.0
RWKV-4-Raven-14B	921	896		25.6	4934	RWKV	Apache 2.0
Alpaca-13B	901	789		48.1	5877	Stanford	Non-commercial
OpenAssistant-Pythia-12B	893	873		27	6380	OpenAssistant	Apache 2.0
ChatGLM-6B	879	884		36.1	4988	Tsinghua	Non-commercial
FastChat-T5-3B	868	759		47.7	4303	LMSYS	Apache 2.0
StableLM-Tuned-Alpha-7B	840	858		24.4	3341	Stability AI	CC-BY-NC-SA-4.0
Dolly-V2-12B	822	746		25.7	3485	Databricks	MIT
LLaMA-13B	799	668		47	2444	Meta	Non-commercial

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.