Arena Math
- What does it measure?
- Preference on maths questions (calculation, worked solutions, proofs). Because people choose rather than checking against an answer key, solutions that are easy to read have an advantage.
- 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. An AI classifier selects prompts that actively apply mathematical concepts — calculations and derivations.
- 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. It measures something different from accuracy tests (AIME, MATH-500): not whether the answer was right, but whether people liked it.
Rank
Top score1535
Models tested104
Scoring basisHigher is better
Last updated2026.11.19
Model release date
Can it solve problems that stump experts across fields?Humanity's Last Exam
Claude Opus 5.5🥇
Claude Opus 5.5🥇Humanity's Last Exam
Claude Opus 5.5🥇Can it solve research-level physics problems?CritPt
Gemini 4 Carbon🥇
Gemini 4 Carbon🥇CritPt
Gemini 4 Carbon🥇
Which AI gives the math answers people prefer most?
Arena Math
- 🥇GoogleGemini 4 Argon57.1%27.1%1535
- 🥈GoogleGemini 4 Neon——1535
- 🥉AnthropicClaude Fable 5.159.1%31.1%1530
- #4AnthropicClaude Fable 555.5%28.6%1521
- #5GoogleGemini 3.8 Flash47.8%18.3%1519
- #6AnthropicClaude Opus 5.561.4%31.7%1517
- #7AnthropicClaude Opus 554.9%29.1%1516
- #8Z.aiGLM 5.3 Flash39.9%15.4%1509
- #9AnthropicClaude Fable 5.2—35.7%1507
- #10MetaMuse Wonder50.1%—1504
- #11GoogleGemini 4 Hydrogen——1504
- #12Z.aiGLM 5.342.3%19.1%1504
- #13AnthropicClaude Opus 4.639.9%—1504
- #14Moonshot AIKimi K346.9%23.4%1502
- #15MetaMuse Spark 1.348.7%26.0%1502
- #16GoogleGemini 3.7 Flash47.9%14.3%1501
- #16GoogleGemini 3.6 Flash40.8%10.6%1501
- #18OpenAIGPT-5.545.8%27.1%1500
- #19OpenAIGPT-6.1 Sol52.9%31.7%1499
- #20AnthropicClaude Sonnet 5.555.0%31.4%1497
- #21DeepSeekDeepSeek V4.1 Flash39.2%14.3%1496
- #22OpenAIGPT-5.6 Sol49.5%32.3%1494
- #23MetaMuse Spark 1.146.2%15.1%1494
- #24OpenAIGPT-5.443.7%23.4%1491
- #25AnthropicClaude Opus 4.742.3%12.0%1490
- #26GoogleGemini 3.1 Pro47.0%17.7%1488
- #27Z.aiGLM 5.241.1%20.9%1488
- #28MetaMuse Spark 1.245.5%17.7%1483
- #29GoogleGemini 3.5 Flash42.7%13.1%1482
- #30XiaomiMiMo-V2.6-Pro49.4%26.6%1482
- #31AlibabaQwen3.7 Max40.5%13.4%1481
- #32OpenAIGPT-6.1 Terra46.2%28.0%1479
- #33Moonshot AIKimi K2.637.5%8.0%1479
- #34XiaomiMiMo V2.5 Pro35.7%4.0%1479
- #35OpenAIGPT-5.6 Terra42.9%30.0%1478
- #36OpenAIGPT-6 Astra54.7%31.7%1477
- #37Moonshot AIKimi K3.1 Pro52.5%24.6%1477
- #38OpenAIGPT-5.6 Luna39.5%20.6%1476
- #39AnthropicClaude Opus 4.848.7%20.9%1475
- #40GoogleGemini 3 Flash36.6%—1475
- #41Z.aiGLM-5.130.1%4.6%1474
- #42StepfunStep-546.5%20.9%1474
- #43AlibabaQwen3.6 Max30.8%—1472
- #44AnthropicClaude Sonnet 541.3%16.9%1472
- #45SpaceXSIGrok 4.542.7%15.4%1471
- #46Moonshot AIKimi K2.530.7%3.1%1470
- #47MetaMuse Spark40.7%11.3%1468
- #48GoogleGemma 4 31B23.6%1.4%1466
- #49AlibabaQwen3.7 Plus35.6%9.1%1465
- #50SpaceXSIGrok 4.20 (Reasoning)34.5%—1464
- #51XiaomiMiMo-V2.6-Flash35.1%12.0%1463
- #52AnthropicClaude Opus 4.530.1%—1463
- #53AnthropicClaude Sonnet 4.633.6%3.1%1461
- #54AlibabaQwen3.6 Plus27.8%2.9%1451
- #55AlibabaQwen3.5 397B A17B29.0%0.9%1451
- #56XiaomiMiMo V2 Pro30.4%—1450
- #57ByteDanceDola Seed 2.0 Pro——1449
- #58SpaceXSIGrok 4.2027.9%—1449
- #59OpenAIGPT-6 Luna38.5%19.4%1449
- #60Mistral AIMistral Large 435.0%11.1%1448
- #61AnthropicClaude Haiku 5.544.4%22.6%1447
- #62SpaceXSIGrok 4.644.1%19.7%1445
- #63OpenAIGPT-6 Sol47.9%30.9%1444
- #64AnthropicClaude Opus 4.112.5%—1444
- #65SpaceXSIGrok 4.743.1%18.0%1443
- #66DeepSeekDeepSeek V4 Pro41.0%18.6%1443
- #67Z.aiGLM-529.3%—1443
- #68NVIDIANemotron 3 Ultra28.4%3.1%1442
- #69GoogleGemini 3.5 Flash-Lite18.8%0.0%1442
- #70GoogleGemini 2.5 Pro22.5%2.0%1439
- #71OpenAIGPT-5.4 Mini28.1%10.0%1439
- #72XiaomiMiMo V2.527.2%3.7%1437
- #73GoogleGemini 3.1 Flash Lite17.2%1.1%1436
- #74OpenAIGPT-528.5%12.6%1435
- #75MeituanLongcat Flash Chat5.8%—1434
- #76BaiduERNIE 5.0 Thinking13.3%—1434
- #77TencentHy333.5%—1429
- #78Mistral AIMistral Medium 3.513.8%0.0%1427
- #79AnthropicClaude Sonnet 4.517.8%1.1%1426
- #80DeepSeekDeepSeek V3.224.6%—1426
- #81DeepSeekDeepSeek V4 Flash38.6%16.6%1424
- #82OpenAIGPT-5.4 Nano28.3%9.3%1424
- #83MiniMaxMiniMax M339.0%3.7%1423
- #84MiniMaxMiniMax M2.729.6%0.6%1423
- #85AnthropicClaude Opus 412.3%0.3%1421
- #86SpaceXSIGrok 4.1 Fast (Reasoning)19.3%—1420
- #87SpaceXSIGrok 4.337.2%8.0%1419
- #88UpstageSolar Pro 429.8%—1416
- #89OpenAIGPT-5 Mini21.5%0.0%1407
- #90GoogleGemini 2.5 Flash12.1%1.1%1405
- #91AnthropicClaude Sonnet 410.7%0.3%1404
- #92AnthropicClaude Haiku 4.510.4%0.0%1402
- #93MiniMaxMiniMax M2.520.5%—1392
- #94Arcee AITrinity Large Thinking15.8%0.9%1383
- #95OpenAIGPT OSS 120B19.6%1.1%1381
- #96OpenAIGPT-4.14.2%—1375
- #97NVIDIANemotron 3 Super20.8%3.1%1370
- #98GoogleGemini 2.5 Flash Lite7.0%—1361
- #99OpenAIGPT-5 Nano9.5%—1348
- #100NVIDIANemotron 3 Nano 30B A3B11.4%—1347
- #101AmazonNova 2 Lite11.6%—1329
- #102IBMGranite 4.1 8B3.8%—1318
- #103MetaLlama 4 Maverick4.9%0.0%1317
- #104MetaLlama 4 Scout3.8%0.0%1310
Half of models ≤ 1462Measured 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$3.301437
49 of the 64 AIs released in the last 60 days have no score here yet · GPT-6.2 Astra, Microsoft-Decision-1.1, GPT-6.2 Sol and 46 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.