AI 리더보드
전체 및 카테고리별 선두 모델을 확인하세요. YouTube와 X에서 볼 수 있는 차트의 기반이 되는 공개 벤치마크입니다. 자체 점수는 공개하지 않으며, 모든 수치는 해당 벤치마크에서 산출된 것으로 날짜와 출처를 명시합니다.
Real people are shown the same prompt answered by two anonymous models side by side and vote for the better answer. Millions of these blind head-to-head votes are fed into a Bradley-Terry statistical model (the successor to the Elo system it started with) which converts win/loss pairs into a single rating per model. A higher rating means people picked that model more often against strong opposition. This snapshot uses the 'style control' variant, which is the site's default: it statistically adjusts for answer length and formatting so a model cannot climb simply by writing longer, prettier replies.
한계: It measures which answer people LIKE, not which answer is CORRECT - a confident, well-written wrong answer can still win a vote. Voters are self-selected volunteers rather than a representative sample, prompts skew toward what that crowd chooses to type, and models with few votes have wide confidence intervals (ci_low/ci_high) that often overlap the models ranked above and below them. Treat small rank gaps as ties.
출처: LMArena (formerly LMSYS Chatbot Arena) ↗ · 공개됨 2026-08-11 · 기반 데이터 ↗ · Dataset released under Creative Commons Attribution 4.0 (CC BY 4.0). Reuse is permitted with attribution - credit LMArena and link to the leaderboard.
| # | 모델 | 조직 | Arena score (human preference) |
|---|---|---|---|
| 1 | claude-opus-5-max | Anthropic | 1555 95% CI 1526 to 1584 426 votes |
| 2 | claude-fable-5 | Anthropic | 1528 95% CI 1510 to 1546 1,043 votes |
| 3 | claude-opus-5-high | Anthropic | 1525 95% CI 1505 to 1546 911 votes |
| 4 | claude-opus-4-6-thinking | Anthropic | 1516 95% CI 1506 to 1526 3,700 votes |
| 5 | gemini-3.6-flash | 1513 95% CI 1490 to 1537 651 votes | |
| 6 | qwen3.8-max | Alibaba | 1513 95% CI 1481 to 1544 340 votes |
| 7 | gemini-3.5-flash-high | 1507 95% CI 1490 to 1524 1,289 votes | |
| 8 | claude-opus-4-6 | Anthropic | 1506 95% CI 1496 to 1516 4,113 votes |
| 9 | claude-opus-4-7-thinking | Anthropic | 1504 95% CI 1492 to 1515 3,022 votes |
| 10 | gpt-5.5 | OpenAI | 1499 95% CI 1487 to 1510 2,853 votes |
막대 그래프는 가시 범위 내에서 작은 차이도 읽기 쉽도록 조정되어 있으며, 0부터 시작하지 않습니다. 신뢰 구간이 겹치는 경우 모델 간 통계적 차이가 없습니다. 상위 그룹은 엄격한 순위가 아닌 하나의 그룹으로 읽어주세요.
이 결과를 읽는 방법, 그리고 왜 서로 다른지
LMArena (formerly LMSYS Chatbot Arena)
Real people are shown the same prompt answered by two anonymous models side by side and vote for the better answer. Millions of these blind head-to-head votes are fed into a Bradley-Terry statistical model (the successor to the Elo system it started with) which converts win/loss pairs into a single rating per model. A higher rating means people picked that model more often against strong opposition. This snapshot uses the 'style control' variant, which is the site's default: it statistically adjusts for answer length and formatting so a model cannot climb simply by writing longer, prettier replies.
주의하세요: It measures which answer people LIKE, not which answer is CORRECT - a confident, well-written wrong answer can still win a vote. Voters are self-selected volunteers rather than a representative sample, prompts skew toward what that crowd chooses to type, and models with few votes have wide confidence intervals (ci_low/ci_high) that often overlap the models ranked above and below them. Treat small rank gaps as ties.
게시됨 2026-08-11 · 실시간 리더보드 ↗ · Data: LMArena leaderboard dataset (CC BY 4.0).
LiveBench
A fixed set of test questions with objectively verifiable answers is run against each model and scored automatically against ground truth - no human voting and no AI judge, so the score is repeatable. This release spans 23 tasks grouped into 7 categories. Each category score is the average of its tasks, and the headline 'global average' is the average of the 7 category scores, so every category counts equally regardless of how many tasks it contains. Scores are percentages: 100 is perfect.
주의하세요: Contamination-LIMITED, not contamination-proof: questions are refreshed from recent sources to reduce the chance a model simply memorised them during training, but that cannot be guaranteed. Scores reflect only these 23 tasks - they say nothing about tone, safety, speed or cost. Many entries are effort/thinking variants of the same underlying model (model_id shows the exact configuration tested), and a variant given more reasoning budget will usually outscore the cheaper default that most people actually use.
게시됨 · 실시간 리더보드 ↗ · Data: LiveBench 2026-06-25 release, livebench.ai.
두 가지 단위는 서로 다릅니다. 하나의 차트로 표시하지 않습니다
인간 선호도 평가와 정답률은 같은 축을 공유할 수 없으므로, 이 페이지에서는 두 지표를 하나의 표에 혼합하지 않습니다. 한 모델이 한쪽에서 1위를 차지하고 다른 쪽에서는 그렇지 않을 수 있는데, 이는 모순이 아니라 해당 모델이 무엇에 강한지를 보여주는 실질적인 신호입니다. 하나는 "사람들이 어떤 답변을 선호했는가?"를 묻고, 다른 하나는 "어떤 답변이 옳았는가?"를 묻습니다.
이 페이지에서 하지 않는 것
어떤 AI를 사용해야 하는지 알려드리지 않습니다. 벤치마크 1위가 항상 최선의 도구는 아닙니다. 가격, 가용성, 연동 기능, 컨텍스트 길이, 그리고 내 지침을 얼마나 잘 따르는지가 보통 점수 차이 몇 점보다 훨씬 중요합니다. 어떤 AI를 사용해야 하나요? →