Arena Coding
- What does it measure?
- Which AI's answers people preferred on coding questions (writing code, fixing bugs, explaining). It is not a test that checks whether the code actually runs — it records which answer the asker found more helpful.
- 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. A prompt is classified as coding automatically when the prompt or answer contains a code block or mentions a programming language.
- 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. Rankings can differ from benchmarks that execute the code (Terminal-Bench, SciCode, etc.).
Rank
Top score1568
Models tested108
Scoring basisHigher is better
Last updated2026.11.19
Model release date
Can AI carry out long development tasks autonomously by operating a computer itself?Terminal-Bench 4.0
Kimi K3.1 Pro🥇
Kimi K3.1 Pro🥇Terminal-Bench 4.0
Kimi K3.1 Pro🥇Can AI test and fix code itself until it completes the given task?LiveBench Agentic Coding v1
DeepSeek V4.1 Flash🥇
DeepSeek V4.1 Flash🥇LiveBench Agentic Coding v1
DeepSeek V4.1 Flash🥇
Which AI do people prefer most for coding answers?
Arena Coding
- 🥇GoogleGemini 4 Carbon—1568—
- 🥈GoogleGemini 4 Neon—1559—
- 🥉GoogleGemini 4 Argon57.1%1552—
- #4AnthropicClaude Fable 542.4%155162.2
- #5MetaMuse Wonder52.0%1550—
- #6AnthropicClaude Opus 5.559.6%154865.3
- #7OpenAIGPT-6 Astra59.6%154857.3
- #8AnthropicClaude Opus 4.6—154749.0
- #9AnthropicClaude Opus 4.7—154650.7
- #10OpenAIGPT-6.1 Sol56.1%154556.8
- #11OpenAIGPT-6.1 Terra55.6%1541—
- #12MetaMuse Spark 1.27.1%154057.6
- #13XiaomiMiMo-V2.6-Pro34.8%1539—
- #14Moonshot AIKimi K312.6%153762.2
- #15MetaMuse Spark 1.333.3%153564.1
- #16MetaMuse Spark 1.16.1%153358.5
- #17OpenAIGPT-5.6 Sol39.9%153356.2
- #18AnthropicClaude Fable 5.155.1%153366.1
- #19Moonshot AIKimi K3.1 Pro67.2%1532—
- #20GoogleGemini 3.8 Flash19.7%153054.2
- #21MetaMuse Spark—1529—
- #22AnthropicClaude Opus 4.821.7%152850.5
- #22AnthropicClaude Sonnet 4.63.0%152842.6
- #24AnthropicClaude Opus 549.0%152865.2
- #25AnthropicClaude Fable 5.2—1527—
- #26Z.aiGLM 5.3 Flash32.8%152656.8
- #27DeepSeekDeepSeek V4.1 Flash26.8%152577.3
- #28Z.aiGLM 5.341.9%152460.9
- #29AlibabaQwen3.7 Max1.5%152443.6
- #30AnthropicClaude Opus 4.5—152339.7
- #30OpenAIGPT-6.2 Astra—1523—
- #32OpenAIGPT-5.4—152153.8
- #33GoogleGemini 3.7 Flash13.6%152158.3
- #33GoogleGemini 3.1 Pro4.0%152144.1
- #35XiaomiMiMo V2.5 Pro0.0%1520—
- #36AnthropicClaude Sonnet 514.1%151959.4
- #37OpenAIGPT-6 Sol43.9%151952.9
- #38GoogleGemini 3.6 Flash7.1%151843.4
- #39OpenAIGPT-5.6 Terra35.4%151855.0
- #40AnthropicClaude Sonnet 5.563.6%151739.3
- #41Moonshot AIKimi K2.60.5%151646.9
- #42Z.aiGLM 5.21.0%151651.8
- #43SpaceXSIGrok 4.510.6%151456.5
- #44ByteDanceDola Seed 2.0 Pro—1514—
- #45GoogleGemini 4 Hydrogen—1514—
- #46Z.aiGLM-5.12.0%1512—
- #46AnthropicClaude Sonnet 4.50.0%1512—
- #48AnthropicClaude Opus 4.1—1511—
- #49SpaceXSIGrok 4.20 (Reasoning)—1511—
- #50AlibabaQwen3.6 Max—1510—
- #51OpenAIGPT-5.514.6%151054.0
- #52XiaomiMiMo-V2.6-Flash22.7%1510—
- #53SpaceXSIGrok 4.621.2%150857.0
- #54SpaceXSIGrok 4.20—1508—
- #55GoogleGemini 3.5 Flash6.6%150649.0
- #56OpenAIGPT-6.2 Sol—1506—
- #57GoogleGemini 3 Flash—1505—
- #58StepfunStep-533.3%1504—
- #59OpenAIGPT-6 Luna12.6%150451.2
- #60OpenAIGPT-5.6 Luna11.6%150348.4
- #60XiaomiMiMo V2 Pro—1503—
- #62GoogleGemma 4 31B0.0%1502—
- #63Moonshot AIKimi K2.5—1502—
- #64GoogleGemini 3.5 Flash-Lite1.0%150245.3
- #65DeepSeekDeepSeek V4 Pro14.1%150142.6
- #65MeituanLongcat Flash Chat—1501—
- #67AlibabaQwen3.7 Plus1.0%1500—
- #68Z.aiGLM-5—1498—
- #69AnthropicClaude Opus 4—1498—
- #70AlibabaQwen3.6 Plus—149641.4
- #71OpenAIGPT-5.4 Mini2.0%149541.7
- #72MiniMaxMiniMax M32.0%149240.7
- #73XiaomiMiMo V2.50.0%1491—
- #74SpaceXSIGrok 4.725.8%149154.0
- #75AlibabaQwen3.5 397B A17B0.0%1491—
- #76BaiduERNIE 5.0 Thinking—1490—
- #77AnthropicClaude Haiku 5.532.8%148751.4
- #78SpaceXSIGrok 4.30.0%148718.5
- #79Mistral AIMistral Large 426.8%148657.2
- #80InklingInkling 2 Small—1485—
- #81DeepSeekDeepSeek V4 Flash12.1%148437.6
- #82AnthropicClaude Haiku 4.50.0%1482—
- #83MiniMaxMiniMax M2.70.0%1480—
- #84Mistral AIMistral Medium 3.50.0%1478—
- #85AnthropicClaude Sonnet 4—1475—
- #86NVIDIANemotron 3 Ultra0.5%147438.7
- #87OpenAIGPT-5—1472—
- #88DeepSeekDeepSeek V3.2—1470—
- #89GoogleGemini 2.5 Pro0.0%1464—
- #90SpaceXSIGrok 4.1 Fast (Reasoning)—1461—
- #91UpstageSolar Pro 40.0%1461—
- #92OpenAIGPT-5.4 Nano0.5%146046.8
- #93GoogleGemini 3.1 Flash Lite0.5%1458—
- #94OpenAIGPT-4.1—1456—
- #95TencentHy30.5%1452—
- #96MiniMaxMiniMax M2.5—1446—
- #97OpenAIGPT-5 Mini0.0%1432—
- #98GoogleGemini 2.5 Flash—1424—
- #99Arcee AITrinity Large Thinking0.5%1412—
- #100NVIDIANemotron 3 Super0.0%1407—
- #101GoogleGemini 2.5 Flash Lite—1398—
- #102AmazonNova 2 Lite—1393—
- #103OpenAIGPT OSS 120B0.0%1389—
- #104OpenAIGPT-5 Nano—1386—
- #105MetaLlama 4 Maverick0.0%1372—
- #106MetaLlama 4 Scout0.0%1362—
- #107NVIDIANemotron 3 Nano 30B A3B0.0%1361—
- #108IBMGranite 4.1 8B—1354—
Half of models ≤ 1507Measured 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$3.301489
45 of the 64 AIs released in the last 60 days have no score here yet · Microsoft-Decision-1.1, d1.2, Qwen4 Max and 42 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.