AI leaderboard
Model mana yang unggul, secara keseluruhan maupun per kategori. Ini adalah benchmark publik yang menjadi dasar grafik yang kamu lihat di YouTube dan X. Kami tidak menerbitkan skor sendiri: setiap angka milik benchmark yang menghasilkannya, diambil pada tanggal tertentu dan dikreditkan ke sumbernya.
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.
Keterbatasan: 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.
Sumber: LiveBench ↗ · diterbitkan · data yang mendasari ↗ · Benchmark code and data are public on GitHub and Hugging Face under the project's own terms; cite LiveBench and link to livebench.ai when reusing scores.
| # | Model | Organisasi | Score out of 100 (objective tasks) |
|---|---|---|---|
| 1 | claude-opus-5-max-effort | Anthropic | 65.2 |
| 2 | Smaug-Agentic | Abacus.AI | 64.7 |
| 3 | Qwen 3.8 Max | Alibaba | 64.7 |
| 4 | Claude Fable 5 Max Effort | Anthropic | 62.2 |
| 5 | Kimi K3 | Moonshot AI | 62.2 |
| 6 | Claude Sonnet 5 xHigh Effort | Anthropic | 59.4 |
| 7 | Muse Spark 1.1 xHigh Effort | Meta | 58.5 |
| 8 | Muse Spark 1.2 xHigh Effort | Meta | 57.6 |
| 9 | Grok 4.5 | xAI | 56.5 |
| 10 | GPT-5.6 Sol Max Effort | OpenAI | 56.2 |
| 11 | GPT-5.6 Terra Max Effort | OpenAI | 55.0 |
| 12 | gpt-5.5-xhigh | OpenAI | 54.0 |
| 13 | gpt-5.4-xhigh | OpenAI | 53.8 |
| 14 | claude-opus-4-7-xhigh-effort | Anthropic | 50.7 |
| 15 | Claude 4.8 Opus Thinking Max Effort | Anthropic | 50.5 |
| 16 | gpt-5.2-2025-12-11-high | OpenAI | 50.2 |
| 17 | gemini-3.5-flash-high | 49.0 | |
| 18 | Claude 4.6 Opus Thinking High Effort | Anthropic | 49.0 |
| 19 | DeepSeek V4 Flash 0731 | DeepSeek | 46.8 |
| 20 | Gemini 3.1 Pro Preview High | 44.1 |
Batang diskalakan pada rentang yang terlihat agar perbedaan kecil tetap mudah dibaca. Batang tidak dimulai dari nol. Di mana confidence interval saling tumpang tindih, model-model tersebut seri secara statistik: baca kelompok teratas sebagai satu kelompok, bukan urutan ketat. Beberapa entri adalah konfigurasi dengan upaya maksimal, yang mendapat skor lebih tinggi dari pengaturan default yang umumnya digunakan orang. Nama model menunjukkan mana yang diuji.
Cara membaca ini, dan mengapa keduanya berbeda
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.
Waspadai: 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.
Diterbitkan 2026-08-11 · live leaderboard ↗ · 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.
Waspadai: 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.
Diterbitkan · live leaderboard ↗ · Data: LiveBench 2026-06-25 release, livebench.ai.
Dua unit yang berbeda, jangan dalam satu grafik
Penilaian preferensi manusia dan skor persentase-benar tidak dapat berbagi satu sumbu, sehingga halaman ini tidak pernah mencampurnya dalam satu tabel. Sebuah model bisa unggul di satu sisi namun tidak di sisi lain, dan itu adalah sinyal nyata tentang keunggulannya, bukan kontradiksi: satu bertanya "jawaban mana yang lebih disukai orang?", yang lain bertanya "jawaban mana yang benar?".
Yang tidak akan dilakukan halaman ini
Ini tidak akan memberi tahu AI mana yang harus digunakan. Pemimpin benchmark sering kali bukan alat yang tepat untuk pekerjaanmu, harga, ketersediaan, integrasi, panjang konteks, dan seberapa baik ia mengikuti instruksi KAMU biasanya lebih penting daripada selisih satu dua poin skor. AI mana yang sebaiknya aku gunakan? →