Specialist leaderboards
The model ranking compares general-purpose models. Systems built for one kind of task are ranked here, each against everything measured on its kind of benchmark. How these boards work.
Voice agents
Hold a spoken conversation in real time: speech in, speech out. 23 ranked systems on 65 benchmarks.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Phonic 1 | Phonic | 1869 ± 115 | 21 | Specialist |
| 2 | Grok Voice 3.0 | xAI | 1740 ± 116 | 14 | Specialist |
| 3 | MiniCPM-o-4.5 | OpenBMB | 1740 ± 138 | 12 | Not ranked |
| 4 | GPT-Live 1 | OpenAI | 1650 ± 141 | 8 | Specialist |
| 5 | Grok Voice Agent | xAI | 1646 ± 142 | 10 | Specialist |
| 6 | Gemini 2.5 Native Audio | 1625 ± 135 | 11 | Specialist | |
| 7 | Grok Voice 2.0 Fast | xAI | 1599 ± 132 | 5 | Specialist |
| 8 | Grok Voice 2.0 | xAI | 1592 ± 97 | 20 | Specialist |
| 9 | GPT Realtime 2.1 | OpenAI | 1577 ± 99 | 13 | Specialist |
| 10 | GPT Realtime 1.5 | OpenAI | 1566 ± 84 | 30 | Specialist |
Speech recognition
Turn recorded speech into text, scored mostly by word error rate. 34 ranked systems on 121 benchmarks; general models lead in 1 of 6 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Sarvam Audio | Sarvam AI | 1925 ± 142 | 17 | Not ranked |
| 2 | Qwen3-ASR-1.7B | Alibaba | 1834 ± 128 | 22 | Specialist |
| 3 | Qwen3 Omni 30B A3B Instruct | Alibaba | 1825 ± 111 | 18 | #401 |
| 4 | parakeet-tdt-0.6B-v3 | NVIDIA | 1706 ± 125 | 20 | Specialist |
| 5 | IndicConformer | AI4Bharat | 1704 ± 126 | 16 | Specialist |
| 6 | Qwen3-Omni-30B-A3B | Alibaba | 1687 ± 113 | 10 | Not ranked |
| 7 | elevenlabs-scribe_v2 | ElevenLabs | 1684 ± 122 | 15 | Specialist |
| 8 | Gemini 2.5 Flash | 1670 ± 128 | 7 | #255 | |
| 9 | Gemini 3 Flash | 1658 ± 103 | 22 | #103 | |
| 10 | Qwen 3.5 Omni Plus | Alibaba | 1650 ± 107 | 31 | #111 |
Image generation
Create and edit images from a text request. 66 ranked systems on 206 benchmarks; general models lead in 2 of 3 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Nano Banana Pro | 2190 ± 92 | 96 | Specialist | |
| 2 | GPT Image 2 | OpenAI | 2083 ± 110 | 38 | Specialist |
| 3 | Nano Banana 2 | 2037 ± 89 | 63 | Specialist | |
| 4 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 1940 ± 122 | 18 | Specialist | |
| 5 | GPT Image 1.5 | OpenAI | 1924 ± 72 | 84 | Specialist |
| 6 | GPT-4o | OpenAI | 1868 ± 132 | 21 | #357 |
| 7 | seedream-lite-5.0 | ByteDance | 1853 ± 91 | 41 | Specialist |
| 8 | Seedream 4.5 | ByteDance | 1847 ± 66 | 84 | Specialist |
| 9 | Seedream 5.0 | ByteDance | 1824 ± 116 | 21 | Specialist |
| 10 | Nano Banana | 1794 ± 76 | 55 | Specialist |
Video generation
Create video from a text request or an image. 21 ranked systems on 89 benchmarks.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Seedance 2.0 | ByteDance | 1879 ± 100 | 32 | Specialist |
| 2 | Sora2 | OpenAI | 1752 ± 135 | 11 | Specialist |
| 3 | Kling 2.6 | Kuaishou | 1737 ± 107 | 19 | Specialist |
| 4 | Kling 3.0 Omni | Kuaishou | 1713 ± 122 | 12 | Specialist |
| 5 | Kling 3.0 | Kuaishou | 1694 ± 123 | 13 | Specialist |
| 6 | Veo 3.1 | 1689 ± 87 | 33 | Specialist | |
| 7 | Veo 3.1 Fast | 1683 ± 90 | 34 | Specialist | |
| 8 | Wan2.7 | Alibaba | 1642 ± 110 | 13 | Specialist |
| 9 | LingBot-World | Other | 1632 ± 119 | 10 | Specialist |
| 10 | Wan2.2 | Alibaba | 1582 ± 131 | 7 | Specialist |
Embeddings and search
Find the relevant documents: embedding, retrieval and reranking models. 18 ranked systems on 126 benchmarks; general models lead in 1 of 3 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Qwen3-Reranker-8B | Alibaba | 1799 ± 126 | 13 | Specialist |
| 2 | Gemini Embedding 001 | 1693 ± 115 | 16 | Specialist | |
| 3 | Qwen3-Reranker-0.6B | Alibaba | 1693 ± 116 | 13 | Specialist |
| 4 | Qwen3-Embedding-8B | Alibaba | 1647 ± 97 | 21 | Specialist |
| 5 | Qwen3-Embedding-4B | Alibaba | 1594 ± 109 | 16 | Specialist |
| 6 | text-embedding-3-large | OpenAI | 1535 ± 107 | 16 | Specialist |
| 7 | Qwen3-Embedding-0.6B | Alibaba | 1495 ± 92 | 21 | Specialist |
| 8 | multilingual-e5-large | Microsoft | 1494 ± 100 | 17 | Not ranked |
| 9 | multilingual-e5-base | Microsoft | 1492 ± 123 | 9 | Not ranked |
| 10 | BGE-M3 | BAAI | 1463 ± 90 | 23 | Specialist |
Robot control
Drive a robot from what its cameras see. 13 ranked systems on 95 benchmarks.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | MolmoAct2 | Allen AI | 1753 ± 155 | 6 | Specialist |
| 2 | pi-0.5 | Physical Intelligence | 1730 ± 92 | 29 | Specialist |
| 3 | LangForce | Other | 1639 ± 144 | 6 | Not ranked |
| 4 | X-VLA | Other | 1609 ± 143 | 5 | Specialist |
| 5 | GR00T-N1.7 | NVIDIA | 1595 ± 141 | 6 | Specialist |
| 6 | pi-0 | Physical Intelligence | 1547 ± 87 | 20 | Specialist |
| 7 | LingBot-VLA | Other | 1533 ± 118 | 11 | Specialist |
| 8 | Motus | Other | 1525 ± 114 | 12 | Not ranked |
| 9 | OpenVLA | Other | 1439 ± 100 | 16 | Specialist |
| 10 | GR00T-N1.6 | NVIDIA | 1437 ± 96 | 22 | Specialist |
Document reading
Turn scans, photos and PDFs into text and structure (OCR). 114 ranked systems on 142 benchmarks; general models lead in 21 of 29 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 1781 ± 119 | 10 | #71 | |
| 2 | Qwen 3 VL Plus | Alibaba | 1781 ± 146 | 6 | Not ranked |
| 3 | Qwen 3.8 27B | Alibaba | 1765 ± 104 | 23 | #123 |
| 4 | Gemini 3 Pro | 1764 ± 106 | 15 | #82 | |
| 5 | Gemini 3.5 Flash | 1755 ± 102 | 15 | #66 | |
| 6 | Gemini 3 Flash | 1749 ± 97 | 17 | #103 | |
| 7 | Gemini 3.6 Flash | 1735 ± 107 | 22 | #57 | |
| 8 | GPT-5.5 | OpenAI | 1721 ± 84 | 20 | #33 |
| 9 | FD-RL | Other | 1721 ± 120 | 17 | Not ranked |
| 10 | Qwen 3.7 Plus | Alibaba | 1717 ± 119 | 6 | #98 |
Computer use
Operate apps and websites through the screen. 58 ranked systems on 115 benchmarks; general models lead in 14 of 19 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | Anthropic | 1804 ± 118 | 9 | #16 |
| 2 | UI-Venus-2-9B | InclusionAI | 1795 ± 122 | 17 | Specialist |
| 3 | Claude Opus 4.8 | Anthropic | 1753 ± 129 | 6 | #39 |
| 4 | Seed 2.0 Pro | ByteDance | 1750 ± 92 | 28 | #122 |
| 5 | Claude Fable 5 | Anthropic | 1748 ± 138 | 5 | #20 |
| 6 | Gemini 3 Pro | 1742 ± 129 | 10 | #82 | |
| 7 | GPT-5.6 Sol | OpenAI | 1737 ± 120 | 6 | #22 |
| 8 | Gemini 3.1 Pro (Preview) | 1736 ± 87 | 36 | #64 | |
| 9 | GPT-5.4 | OpenAI | 1708 ± 98 | 27 | #83 |
| 10 | Claude Opus 4.6 | Anthropic | 1658 ± 90 | 33 | #67 |
Deep research
Search the web and write a long research report. 25 ranked systems on 86 benchmarks; general models lead in 1 of 7 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | GPT-5.4 | OpenAI | 1805 ± 162 | 9 | #83 |
| 2 | Gemini 3 Pro Deep Research | 1780 ± 132 | 9 | Specialist | |
| 3 | Gemini 2.5 Pro Deep Research | 1778 ± 125 | 10 | Specialist | |
| 4 | O3 Deep Research | OpenAI | 1615 ± 83 | 25 | Specialist |
| 5 | Deep Research | OpenAI | 1562 ± 93 | 19 | Specialist |
| 6 | Gemini 3.1 Pro Deep Research | 1554 ± 116 | 12 | Specialist | |
| 7 | Gemini Deep Research | 1546 ± 147 | 5 | Specialist | |
| 8 | Gemini 3.1 Pro (Preview) | 1533 ± 137 | 9 | #64 | |
| 9 | Claude Opus 4.6 | Anthropic | 1524 ± 114 | 12 | #67 |
| 10 | Sonar Deep Research | Perplexity | 1519 ± 100 | 18 | Specialist |
Judges and reward models
Score or compare other models' answers. 166 ranked systems on 170 benchmarks; general models lead in 13 of 20 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Gemini 3 Flash | 1936 ± 105 | 38 | #103 | |
| 2 | GPT-5.5 | OpenAI | 1856 ± 108 | 33 | #33 |
| 3 | Gemini 3.1 Pro (Preview) | 1832 ± 119 | 13 | #64 | |
| 4 | Qwen 3.5 35B A3B | Alibaba | 1814 ± 133 | 8 | #268 |
| 5 | Qwen 3.5 397B A17B | Alibaba | 1801 ± 134 | 9 | #154 |
| 6 | Kimi K2.5 | Moonshot | 1798 ± 134 | 9 | #152 |
| 7 | Kimi K2.6 | Moonshot | 1796 ± 133 | 9 | #113 |
| 8 | Skywork-Reward-Gemma-2-27B | Skywork | 1793 ± 122 | 13 | Specialist |
| 9 | GLM-5 FP8 | Zhipu | 1792 ± 133 | 9 | #147 |
| 10 | DeepSeek V3.2 Speciale | DeepSeek | 1791 ± 133 | 9 | #138 |
Audio understanding
Answer questions about speech, music and other sounds. 44 ranked systems on 283 benchmarks; general models lead in 17 of 21 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 1894 ± 85 | 48 | #64 | |
| 2 | Qwen 3.5 Omni Plus | Alibaba | 1889 ± 110 | 22 | #111 |
| 3 | Gemini 3.5 Flash | 1860 ± 135 | 14 | #66 | |
| 4 | Gemini 3 Flash | 1815 ± 133 | 15 | #103 | |
| 5 | Gemini 2.5 Pro | 1757 ± 102 | 13 | #159 | |
| 6 | GPT Realtime 2 | OpenAI | 1709 ± 110 | 13 | Voice agent |
| 7 | Gemini 2.0 Flash | 1671 ± 76 | 51 | #351 | |
| 8 | Gemini 2.5 Flash | 1667 ± 77 | 42 | #255 | |
| 9 | Qwen3-Omni-30B-A3B | Alibaba | 1660 ± 75 | 49 | Not ranked |
| 10 | GPT-4o Audio | OpenAI | 1633 ± 62 | 76 | Not ranked |
Translation
Translate text between languages. 29 ranked systems on 52 benchmarks; general models lead in 7 of 9 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 1793 ± 116 | 10 | #64 | |
| 2 | Hy-MT2-30B-A3B | Tencent | 1757 ± 113 | 10 | Specialist |
| 3 | DeepSeek V3 | DeepSeek | 1725 ± 125 | 11 | #329 |
| 4 | GPT-4o | OpenAI | 1707 ± 95 | 36 | #357 |
| 5 | DeepSeek V4 Pro | DeepSeek | 1705 ± 130 | 5 | #112 |
| 6 | Hy-MT2-7B | Tencent | 1685 ± 116 | 9 | Specialist |
| 7 | Gemma 4 31B | 1654 ± 131 | 6 | #212 | |
| 8 | Qwen 3.6 35B A3B | Alibaba | 1649 ± 121 | 8 | #238 |
| 9 | Gemma 4 26B A4B | 1636 ± 141 | 5 | #360 | |
| 10 | Qwen 3.5 27B | Alibaba | 1631 ± 145 | 5 | #192 |
Formal proofs
Write proofs a computer can check, in Lean. 10 ranked systems on 18 benchmarks; general models lead in 4 of 5 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 1753 ± 148 | 6 | #64 | |
| 2 | GPT-5.1 | OpenAI | 1711 ± 154 | 6 | #145 |
| 3 | Goedel-Prover-V2-32B | Princeton University | 1621 ± 118 | 9 | Specialist |
| 4 | Goedel-Formalizer-V2-8B | Princeton University | 1537 ± 134 | 6 | Specialist |
| 5 | GPT-OSS-120B | OpenAI | 1524 ± 130 | 5 | #346 |
| 6 | Kimina-Prover-72B | Moonshot | 1495 ± 134 | 6 | Specialist |
| 7 | DeepSeek-Prover-V2-7B | DeepSeek | 1467 ± 132 | 5 | Specialist |
| 8 | Goedel-Prover-V2-8B | Princeton University | 1464 ± 133 | 5 | Specialist |
| 9 | Qwen 3 235B A22B 2507 (Thinking) | Alibaba | 1260 ± 150 | 6 | #263 |
| 10 | Kimina-Autoformalizer-7B | Moonshot | 1229 ± 148 | 6 | Specialist |
Safety classifiers
Flag harmful requests and replies. 23 ranked systems on 63 benchmarks; general models lead in 8 of 9 sources testing both.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | GPT-5.4 | OpenAI | 1792 ± 117 | 16 | #83 |
| 2 | Gemma 3 27B | 1662 ± 134 | 7 | #606 | |
| 3 | GPT-4o | OpenAI | 1649 ± 105 | 27 | #357 |
| 4 | Qwen 3 8B | Alibaba | 1597 ± 108 | 26 | #678 |
| 5 | PolyGuard-Qwen | Other | 1593 ± 139 | 8 | Specialist |
| 6 | GPT-OSS-20B | OpenAI | 1518 ± 120 | 9 | #523 |
| 7 | Llama-3.1-Nemotron-Safety-Guard-8B-v3 | NVIDIA | 1514 ± 134 | 8 | Specialist |
| 8 | gpt-oss-safeguard-20B | OpenAI | 1464 ± 90 | 20 | Specialist |
| 9 | Qwen3Guard-Gen-0.6B | Alibaba | 1447 ± 139 | 7 | Specialist |
| 10 | Qwen3Guard-8B | Alibaba | 1446 ± 129 | 5 | Specialist |
Forecasting
Predict real-world events before they happen. 23 ranked systems on 19 benchmarks.
| Rank | System | Developer | Board rating | Benchmarks | Main ranking |
|---|---|---|---|---|---|
| 1 | GPT-5.4 | OpenAI | 1707 ± 132 | 7 | #83 |
| 2 | Gemini 3.1 Pro (Preview) | 1629 ± 99 | 14 | #64 | |
| 3 | GPT-4.1 | OpenAI | 1578 ± 117 | 10 | #258 |
| 4 | Kimi K3 | Moonshot | 1558 ± 128 | 7 | #35 |
| 5 | GPT-5 | OpenAI | 1553 ± 127 | 7 | #105 |
| 6 | O4 Mini (2025-04-16) | OpenAI | 1538 ± 115 | 10 | #183 |
| 7 | O3 (2025-04-16) | OpenAI | 1536 ± 115 | 10 | #125 |
| 8 | Grok 4.20 | xAI | 1535 ± 123 | 7 | #179 |
| 9 | Claude Sonnet 4.5 | Anthropic | 1531 ± 126 | 7 | #148 |
| 10 | Grok 4.1 Fast | xAI | 1529 ± 127 | 7 | #227 |
Interactive version: theaggregate.ai/specialists · How It Works · Data refreshed daily, snapshot 2026-11-05.