Arena Multi-turn
- What does it measure?
- Preference when a conversation continues over several turns rather than a single exchange — whether the AI remembers earlier context and stays consistent. Close to the real chat experience.
- Who evaluates it, and how?
- On Arena Intelligence, a user enters a question, two anonymous AIs answer, and the user votes for the better one. Over 82 million votes are aggregated with a statistical model (Bradley-Terry) into an Elo-style score, and this table takes the 'style-controlled' version from the official dataset, which statistically removes the bias toward long, nicely formatted answers. Only conversations with two or more exchanges are counted.
- Reading the score
- This is a head-to-head record, not an exam score (accuracy %). It is computed from the probability of being preferred over other AIs, so there is no maximum, and when a new AI enters, existing scores shift slightly. Typical values are 1,000–1,500; a 100-point gap means the leader is preferred about 64% of the time, a 200-point gap about 76%. Correctness is not counted — a wrong but friendly answer can win — so if accuracy matters, also look at accuracy tests (GPQA, AIME, etc.). Gaps of a few dozen points are effectively the same tier even if the rank differs. Useful for separating AIs that only give a good first answer from those that stay coherent as the conversation grows.
Rank
Top score1565
Models tested108
Scoring basisHigher is better
Last updated2026.11.19
Model release date
Which AI gives the writing and literature answers people prefer most?Arena Writing, Literature & Language
Gemini 4 Carbon🥇
Gemini 4 Carbon🥇Arena Writing, Literature & Language
Gemini 4 Carbon🥇Can it read people's feelings and respond well in hard conversations?EQ-Bench 4
Claude Opus 5🥇
Claude Opus 5🥇EQ-Bench 4
Claude Opus 5🥇
Which AI do people prefer most in long conversations?
Arena Multi-turn
- 🥇GoogleGemini 4 Carbon15271565—
- 🥈GoogleGemini 4 Argon152115421236
- 🥉MetaMuse Wonder149715221325
- #4GoogleGemini 4 Neon15051520—
- #5AnthropicClaude Fable 5150915171313
- #6AnthropicClaude Opus 4.7148315141293
- #7AnthropicClaude Opus 4.6148915091210
- #8MetaMuse Spark 1.214631508—
- #9GoogleGemini 3.8 Flash14911504—
- #10Moonshot AIKimi K3147515021319
- #11AnthropicClaude Opus 5.5151615011340
- #12MetaMuse Spark 1.1145814971244
- #13GoogleGemini 3.1 Pro148214951132
- #14MetaMuse Spark 1.314801494—
- #15AnthropicClaude Opus 4.8146814931268
- #16MetaMuse Spark14631493—
- #17OpenAIGPT-5.4146314931258
- #18OpenAIGPT-5.6 Sol148314911236
- #19GoogleGemini 3.7 Flash14911490—
- #20GoogleGemini 3.6 Flash14741490—
- #21OpenAIGPT-6.1 Sol147014871272
- #22AnthropicClaude Opus 5148114861365
- #23Z.aiGLM 5.314671485—
- #24AnthropicClaude Fable 5.215111485—
- #25OpenAIGPT-5.5146614841295
- #26AnthropicClaude Opus 4.514661484—
- #27AnthropicClaude Fable 5.114991483—
- #28SpaceXSIGrok 4.2014521483—
- #29GoogleGemini 3 Flash14611482—
- #30OpenAIGPT-6 Astra14631481—
- #31GoogleGemini 4 Hydrogen14851481—
- #32AnthropicClaude Sonnet 4.6145514801197
- #33AlibabaQwen3.7 Max145814801100
- #33XiaomiMiMo V2.5 Pro144914801198
- #35GoogleGemini 3.5 Flash147214791077
- #36Z.aiGLM 5.3 Flash14521479—
- #37SpaceXSIGrok 4.20 (Reasoning)14491479—
- #38Z.aiGLM-5.114531478—
- #39AnthropicClaude Sonnet 4.514541477—
- #40OpenAIGPT-6.1 Terra14571477—
- #41Z.aiGLM 5.2145914771208
- #42DeepSeekDeepSeek V4 Pro145114751156
- #43Moonshot AIKimi K3.1 Pro14751475—
- #44OpenAIGPT-5.6 Terra144714741220
- #45AnthropicClaude Sonnet 5145114731221
- #46OpenAIGPT-6.2 Astra14811472—
- #47SpaceXSIGrok 4.514581472—
- #48Z.aiGLM-514471471—
- #49AnthropicClaude Opus 4.114461471—
- #50AnthropicClaude Sonnet 5.514661470—
- #50DeepSeekDeepSeek V4.1 Flash14541470—
- #52OpenAIGPT-6 Sol14381467—
- #53GoogleGemini 3.5 Flash-Lite14431467—
- #54AlibabaQwen3.6 Max14461467—
- #55XiaomiMiMo V2 Pro14331466—
- #56GoogleGemma 4 31B143414651110
- #56OpenAIGPT-5.4 Mini14231465—
- #58OpenAIGPT-6.2 Sol14581465—
- #59AlibabaQwen3.7 Plus14361464—
- #60XiaomiMiMo-V2.6-Pro14581463—
- #61Moonshot AIKimi K2.6144114621191
- #62OpenAIGPT-5.6 Luna143314591146
- #63DeepSeekDeepSeek V4 Flash14161453—
- #64Moonshot AIKimi K2.514261452—
- #64AlibabaQwen3.5 397B A17B14211452—
- #66XiaomiMiMo V2.514111450—
- #67SpaceXSIGrok 4.614501450—
- #68ByteDanceDola Seed 2.0 Pro14101449—
- #69GoogleGemini 2.5 Pro14451448—
- #70SpaceXSIGrok 4.3142414471065
- #71MiniMaxMiniMax M3141014471140
- #72XiaomiMiMo-V2.6-Flash14161446—
- #73AlibabaQwen3.6 Plus14231445—
- #74StepfunStep-514321445—
- #75OpenAIGPT-6 Luna14241444—
- #76BaiduERNIE 5.0 Thinking14191441—
- #77AnthropicClaude Opus 414321439—
- #78GoogleGemini 3.1 Flash Lite14211436—
- #79Mistral AIMistral Large 413791433—
- #80Mistral AIMistral Medium 3.514041432983
- #81SpaceXSIGrok 4.714291432—
- #82MeituanLongcat Flash Chat13921431—
- #83OpenAIGPT-4.114051430—
- #84DeepSeekDeepSeek V3.214101428—
- #85MiniMaxMiniMax M2.713861426—
- #86AnthropicClaude Haiku 4.5139914251053
- #87AnthropicClaude Sonnet 413991421—
- #87OpenAIGPT-513961421—
- #89TencentHy313781416—
- #90SpaceXSIGrok 4.1 Fast (Reasoning)14081414—
- #91OpenAIGPT-5.4 Nano13621413—
- #92InklingInkling 2 Small13821413—
- #93AnthropicClaude Haiku 5.513881411—
- #94GoogleGemini 2.5 Flash14021403—
- #95NVIDIANemotron 3 Ultra13931397—
- #96MiniMaxMiniMax M2.513691397—
- #97UpstageSolar Pro 413421388—
- #98GoogleGemini 2.5 Flash Lite13721373—
- #99OpenAIGPT-5 Mini13511373—
- #100NVIDIANemotron 3 Super13241350—
- #101Arcee AITrinity Large Thinking13431346—
- #102MetaLlama 4 Maverick13121326—
- #103OpenAIGPT OSS 120B13101326—
- #104OpenAIGPT-5 Nano12881326—
- #105AmazonNova 2 Lite12981322—
- #106MetaLlama 4 Scout13041319—
- #107NVIDIANemotron 3 Nano 30B A3B12661289—
- #108IBMGranite 4.1 8B12761281—
Half of models ≤ 1467Measured by: User preference votes
Scores from an earlier edition
These scores are not in the latest leaderboard (2026.11.19) — the model was dropped by the source or has not been refreshed yet. The last score we received is shown for reference, unranked.
- as of 2026.11.09$3.301434
44 of the 59 AIs released in the last 60 days have no score here yet · Microsoft-Decision-1.1, d1.2, d1.2-ultra and 41 more
It’s a preference vote, not a correctness check. A wrong but friendly answer can win.
Which questions got asked depends on who voted — they may differ from your use.
Scores labelled 'Self-reported' were published by the AI maker itself, not measured by a third party.