Chatbot Arena

Attribution LMSYS May 5, 2025

This leaderboard is based on the following benchmarks.

  • Chatbot Arena - a crowdsourced, randomized battle platform for large language models (LLMs). We use 2.9M+ 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

ModelArena EloMMLULicense
DeepSeek DeepSeek-V3-0324137288.5MIT
DeepSeek DeepSeek-R1135890.8MIT
Qwen Qwen3-235B-A22B134288.5Apache 2.0
Gemini Gemma-3-27B-it1341Gemma
Gemini Gemma-3-12B-it1321Gemma
Qwen QwQ-32B1313Apache 2.0
Nvidia Llama-3.3-Nemotron-Super-49B-v1129686Nvidia

Full Leaderboard
ModelArena EloCodingVisionArena HardMMLUVotesOrganizationLicense
🥇 Gemini-2.5-Pro-Exp-03-2514371430132896.412541GoogleProprietary
🥇 o3-2025-04-161411142613055714OpenAIProprietary
🥇 ChatGPT-4o-latest (2025-03-26)14081407131110290OpenAIProprietary
🥇 Grok-3-Preview-02-241402140892.714845xAIProprietary
🥇 GPT-4.5-Preview13981400125615276OpenAIProprietary
🥇 Gemini-2.5-Flash-Preview-04-171393140112745303GoogleProprietary
🥈 Gemini-2.0-Pro-Exp-02-0513801379124020120GoogleProprietary
🥈 Gemini-2.0-Flash-Thinking-Exp-01-2113801364127826892GoogleProprietary
🥈 DeepSeek-V3-03241372139388.58127DeepSeekMIT
🥈 GPT-4.1-2025-04-141367137212824432OpenAIProprietary
🥈 DeepSeek-R11358136593.290.817957DeepSeekMIT
🥈 Gemini-2.0-Flash-00113551353122424289GoogleProprietary
🥈 o4-mini-2025-04-161352137212644412OpenAIProprietary
🥈 o1-2024-12-1713501358123192.191.829042OpenAIProprietary
🥈 Qwen3-235B-A22B1342136595.688.52680AlibabaApache 2.0
🥈 Qwen2.5-Max1341134223169AlibabaProprietary
🥈 Gemma-3-27B-it1341131311801GoogleGemma
🥈 o3-mini-high1325136219405OpenAIProprietary
🥈 GPT-4.1-mini-2025-04-141323136112334293OpenAIProprietary
🥉 Gemma-3-12B-it132112862466GoogleGemma
🥉 DeepSeek-V31318132085.588.522833DeepSeekDeepSeek
🥉 QwQ-32B131313179719AlibabaApache 2.0
🥉 Gemini-2.0-Flash-Lite13121320115924427GoogleProprietary
🥉 GLM-4-Plus-0111131112906027ZhipuProprietary
🥉 Qwen-Plus-0125131013206062AlibabaProprietary
🥉 o3-mini1305134924925OpenAIProprietary
🥉 Command A (03-2025)130513109885CohereCC-BY-NC-4.0
🥉 Step-2-16K-Exp130512955126StepFunProprietary
🥉 o1-mini130413539254952OpenAIProprietary
🥉 Hunyuan-TurboS-20250226130213262450TencentProprietary
🥉 Gemini-1.5-Pro-00213021291122258642GoogleProprietary
🥉 Claude 3.7 Sonnet (thinking-32k)1301133911345AnthropicProprietary
🥉 Hunyuan-Turbo-0110129613152515TencentProprietary
🥉 Llama-3.3-Nemotron-Super-49B-v112961299862368NvidiaNvidia
🥉 Claude 3.7 Sonnet12911328120816696AnthropicProprietary
🥉 Grok-2-08-131288128287.567085xAIProprietary
🥉 Yi-Lightning1287130281.52897001 AIProprietary
🥉 GPT-4o-2024-05-1312851293120679.2188.7117742OpenAIProprietary
🥉 Claude 3.5 Sonnet (20241022)12831326118385.288.765446AnthropicProprietary
Deepseek-v2.5-1210127912977243DeepSeekDeepSeek
Athene-v2-Chat-72B127513008526078NexusFlowNexusFlow
Hunyuan-Large-2025-02-10127212933858TencentProprietary
GPT-4o-mini-2024-07-1812721283112474.948271373OpenAIProprietary
Gemma-3-4B-it127212542773GoogleGemma
GPT-4.1-nano-2025-04-141271128811234452OpenAIProprietary
Gemini-1.5-Flash-00212711254120637025GoogleProprietary
Llama-4-Maverick-17B-128E-Instruct127012867121MetaLlama 4
Llama-3.1-405B-Instruct-bf161269128088.643791MetaLlama 3.1
Llama-3.1-Nemotron-70B-Instruct1269127184.97577NvidiaLlama 3.1
Llama-3.1-405B-Instruct-fp81267127669.388.663035MetaLlama 3.1
Grok-2-Mini-08-131266126255435xAIProprietary
Yi-Lightning-lite126412671707101 AIProprietary
Hunyuan-Standard-2025-02-10126112694017TencentProprietary
Qwen2.5-72B-Instruct125712837841515AlibabaQwen
Llama-3.3-70B-Instruct1257125838102MetaLlama-3.3
GPT-4-Turbo-2024-04-0912561263115182.63102148OpenAIProprietary
Mistral-Large-24071251126970.4248216MistralMistral Research
GPT-4-1106-preview12501253103748OpenAIProprietary
Athene-70B1250125377.620577NexusFlowCC-BY-NC-4.0
Mistral-Large-24111249126529655MistralMRL
Llama-3.1-70B-Instruct1248125155.738658643MetaLlama 3.1
Claude 3 Opus12471250107660.3686.8202653AnthropicProprietary
Amazon Nova Pro 1.012451262104425484AmazonProprietary
GPT-4-0125-preview1245124377.9697072OpenAIProprietary
Llama-3.1-Tulu-3-70B124412333009Ai2Llama 3.1
Yi-Large-preview1240124571.485163701 AIProprietary
Claude 3.5 Haiku (20241022)1237126535694AnthropicPropretary
Reka-Core-20240904123512217946Reka AIProprietary
Reka-Core-202407221231120813280Reka AIProprietary
Qwen-Plus-08281227124514624AlibabaProprietary
Gemini-1.5-Flash-00112271232107249.6178.965665GoogleProprietary
Jamba-1.5-Large1222122781.29127AI21 LabsJamba Open
Deepseek-v2-API-06281220124219510DeepSeek AIDeepSeek
Gemma-2-27B-it1220120957.5179526GoogleGemma license
Qwen2.5-Coder-32B-Instruct121712615734AlibabaApache 2.0
Amazon Nova Lite 1.012171235106120654AmazonProprietary
Mistral-Small-24B-Instruct-25011217123115317MistralApache 2.0
Gemma-2-9B-it-SimPO1216119610551PrincetonMIT
Command R+ (08-2024)1215118110538CohereCC-BY-NC-4.0
Deepseek-Coder-v2-07241214126662.311726DeepSeekProprietary
Yi-Large1213122063.71662401 AIProprietary
Gemini-1.5-Flash-8B-00112131208110637696GoogleProprietary
Llama-3.1-Nemotron-51B-Instruct121112113888NvidiaLlama 3.1
Nemotron-4-340B-Instruct1209119820615NvidiaNVIDIA Open Model
Aya-Expanse-32B1209119328753CohereCC-BY-NC-4.0
GLM-4-05201207121663.8410221Zhipu AIProprietary
Llama-3-70B-Instruct1207119946.5782163650MetaLlama 3
Reka-Flash-20240904120611918136Reka AIProprietary
Phi-41205122125224MicrosoftMIT
OLMo-2-0325-32B-Instruct120511972775Allen AIApache-2.0
Claude 3 Sonnet12011213104846.879113062AnthropicProprietary
Reka-Flash-202407221201118713729Reka AIProprietary
Reka-Core-2024050111991190101583.262563Reka AIProprietary
Amazon Nova Micro 1.01198121020650AmazonProprietary
Gemma-2-9B-it1192117357212GoogleGemma license
Command R+ (04-2024)1190116433.0780856CohereCC-BY-NC-4.0
Hunyuan-Standard-256K118912272900TencentProprietary
Qwen2-72B-Instruct1187118746.8684.238871AlibabaQianwen LICENSE
GPT-4-0314118611955086.455974OpenAIProprietary
Llama-3.1-Tulu-3-8B118511793074Ai2Llama 3.1
GLM-4-01161183119155.727578Zhipu AIProprietary
Qwen-Max-04281183118925697AlibabaProprietary
Ministral-8B-2410118212015109MistralMRL
Aya-Expanse-8B1180116510397CohereCC-BY-NC-4.0
Command R (08-2024)1180116110850CohereCC-BY-NC-4.0
Claude 3 Haiku11791189100041.4775.2122319AnthropicProprietary
DeepSeek-Coder-V2-Instruct1178123915754DeepSeek AIDeepSeek License
Llama-3.1-8B-Instruct1176118621.347352581MetaLlama 3.1
Jamba-1.5-Mini1176118169.79272AI21 LabsJamba Open
GPT-4-06131163116737.991624OpenAIProprietary
Qwen1.5-110B-Chat1161117580.427445AlibabaQianwen LICENSE
Mistral-Large-24021157117037.7181.264909MistralProprietary
Yi-1.5-34B-Chat1157116276.82513401 AIApache-2.0
Reka-Flash-21B-online1156114716029Reka AIProprietary
Llama-3-8B-Instruct1152114620.5668.4109100MetaLlama 3
InternLM2.5-20B-chat1149115810593InternLMOther
Claude-1114911367721150AnthropicProprietary
Command R (04-2024)1149112317.0256388CohereCC-BY-NC-4.0
Mixtral-8x22b-Instruct-v0.11148115336.3677.853753MistralApache 2.0
Mistral Medium1148115231.975.335567MistralProprietary
Qwen1.5-72B-Chat1147116036.1277.540660AlibabaQianwen LICENSE
Reka-Flash-21B1147114173.525804Reka AIProprietary
Gemma-2-2b-it1144110751.348893GoogleGemma license
Granite-3.1-8B-Instruct114311733294IBMApache 2.0
Claude-2.01132113523.9978.512759AnthropicProprietary
Gemini-1.0-Pro-0011131110371.818802GoogleProprietary
Zephyr-ORPO-141b-A35b-v0.1112711244855HuggingFaceApache 2.0
Qwen1.5-32B-Chat1125114973.422766AlibabaQianwen LICENSE
Mistral-Next1124113227.3712376MistralProprietary
Phi-3-Medium-4k-Instruct1123112533.377826112MicrosoftMIT
Granite-3.1-2B-Instruct111911473380IBMApache 2.0
Starling-LM-7B-beta1119112923.0116669NexusflowApache-2.0
Claude-2.11118113222.7737697AnthropicProprietary
GPT-3.5-Turbo-06131117113524.8238955OpenAIProprietary
Mixtral-8x7B-Instruct-v0.11114111423.470.676141MistralApache 2.0
Claude-Instant-11111110973.420626AnthropicProprietary
Yi-34B-Chat1111110623.1573.51591801 AIYi License
Gemini Pro1111109117.871.86556GoogleProprietary
Qwen1.5-14B-Chat1109112667.618688AlibabaQianwen LICENSE
GPT-3.5-Turbo-03141107111518.05705641OpenAIProprietary
GPT-3.5-Turbo-01251106112423.3468858OpenAIProprietary
WizardLM-70B-v1.01106107163.78385MicrosoftLlama 2
DBRX-Instruct-Preview1103111824.6373.733738DatabricksDBRX LICENSE
Llama-3.2-3B-Instruct110310808395MetaLlama 3.2
Phi-3-Small-8k-Instruct1102110729.7775.718473MicrosoftMIT
Tulu-2-DPO-70B1099109314.996659AllenAI/UWAI2 ImpACT Low-risk
Granite-3.0-8B-Instruct109310977002IBMApache 2.0
Llama-2-70B-chat1093107211.556339609MetaLlama 2
OpenChat-3.5-01061092110265.812991OpenChatApache-2.0
Vicuna-33B109110678.6359.222938LMSYSNon-commercial
Snowflake Arctic Instruct1090107717.6167.334177SnowflakeApache 2.0
Starling-LM-7B-alpha1088108012.863.910416UC BerkeleyCC-BY-NC-4.0
Gemma-1.1-7B-it1084108412.0964.325066GoogleGemma license
Nous-Hermes-2-Mixtral-8x7B-DPO108410793836NousResearchApache-2.0
NV-Llama2-70B-SteerLM-Chat1080102368.53635NvidiaLlama 2
pplx-70B-online107810286898Perplexity AIProprietary
DeepSeek-LLM-67B-Chat1077107971.34988DeepSeek AIDeepSeek License
OpenChat-3.51076105464.38107OpenChatApache-2.0
Granite-3.0-2B-Instruct107410887187IBMApache 2.0
OpenHermes-2.5-Mistral-7B107410585087NousResearchApache-2.0
Mistral-7B-Instruct-v0.21072107312.5720061MistralApache-2.0
Phi-3-Mini-4K-Instruct-June-241071108270.912803MicrosoftMIT
Qwen1.5-7B-Chat10701089614873AlibabaQianwen LICENSE
GPT-3.5-Turbo-11061068109518.8717038OpenAIProprietary
Phi-3-Mini-4k-Instruct1066108668.821091MicrosoftMIT
Llama-2-13b-chat1063105153.619715MetaLlama 2
Dolphin-2.2.1-Mistral-7B106310241713Cognitive ComputationsApache-2.0
SOLAR-10.7B-Instruct-v1.01062104766.24288Upstage AICC-BY-NC-4.0
WizardLM-13b-v1.21059102652.77176MicrosoftLlama 2
Llama-3.2-1B-Instruct105410468519MetaLlama 3.2
Qwen2.5-VL-32B-Instruct1212AlibabaApache 2.0
Step-1o-Vision-32k (highres)1187StepFunProprietary
Qwen2.5-VL-72B-Instruct1171AlibabaQwen
Pixtral-Large-24111154MistralMRL
Qwen-VL-Max-11191128AlibabaProprietary
Qwen2-VL-72b-Instruct1111AlibabaQwen
Step-1V-32K1111StepFunProprietary
Molmo-72B-09241076AI2Apache 2.0
Pixtral-12B-24091072MistralApache 2.0
Llama-3.2-90B-Vision-Instruct1070MetaLlama 3.2
Aya-Vision-8B1069CohereCC-BY-NC-4.0
InternVL2-26B1067OpenGVLabMIT
Hunyuan-Standard-Vision-2024-12-311066TencentProprietary
Aya-Vision-32B1060CohereCC-BY-NC-4.0
Qwen2-VL-7B-Instruct1054AliabaApache 2.0
Yi-Vision104501 AIProprietary
Llama-3.2-11B-Vision-Instruct1032MetaLlama 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 RA and player B a rating of RB, the probability of player A winning is

{\displaystyle E_{\mathsf {A}}={\frac {1}{1+10^{(R_{\mathsf {B}}-R_{\mathsf {A}})/400}}}~.}

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.

{\displaystyle R_{\mathsf {A}}'=R_{\mathsf {A}}+K\cdot (S_{\mathsf {A}}-E_{\mathsf {A}})~.}

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.