数据来源：Artificial Analysis 模型榜单 · 更新时间：2026-09-30 14:33:01
默认按 AA 指数降序排列；点击表头可切换排序；表头冻结，滚轮查看更多行。
| 排名 ↕ | 模型名称 ↕ | 生产厂商 ↕ | 类型 ↕ | AA 指数 ↕ | 任务平均成本 ($/task) ↕ | 上下文长度 (tokens) ↕ | 输出速率 (tok/s) ↕ | 首字延迟 (s) ↕ | 总响应时间 (s) ↕ |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (max with fallback) | Anthropic | 闭源 | 58 | 5.980 | 1M | 92.0 | 694.0 | 699.4 |
| 2 | Claude Opus 5.5 (xhigh with fallback) | Anthropic | 闭源 | 56 | 3.460 | 1M | 80.0 | 129.8 | 136.0 |
| 3 | Claude Sonnet 5.5 (max with fallback) | Anthropic | 闭源 | 56 | 7.600 | 1M | 139.0 | 407.2 | 410.8 |
| 4 | Claude Opus 5.5 (high with fallback) | Anthropic | 闭源 | 54 | 1.820 | 1M | 74.0 | 33.2 | 40.0 |
| 5 | Claude Fable 5.1 (max with fallback) | Anthropic | 闭源 | 53 | 7.630 | 1M | 69.0 | 276.4 | 283.6 |
| 6 | Claude Fable 5.1 (xhigh with fallback) | Anthropic | 闭源 | 53 | 5.980 | 1M | 61.0 | 108.0 | 116.2 |
| 7 | GPT-6 Astra (max) | OpenAI | 闭源 | 53 | 3.260 | 1M | 55.0 | 291.0 | 300.1 |
| 8 | GPT-6 Astra (xhigh) | OpenAI | 闭源 | 52 | 2.310 | 1M | 49.0 | 126.2 | 136.4 |
| 9 | Claude Sonnet 5.5 (xhigh with fallback) | Anthropic | 闭源 | 52 | 2.740 | 1M | 103.0 | 33.5 | 38.4 |
| 10 | GPT-6.1 Sol (max) | OpenAI | 闭源 | 52 | 0.720 | 1M | 69.0 | 205.9 | 213.1 |
| 11 | Claude Opus 5.5 (medium with fallback) | Anthropic | 闭源 | 51 | 1.340 | 1M | 73.0 | 24.2 | 31.0 |
| 12 | Claude Fable 5.1 (high with fallback) | Anthropic | 闭源 | 51 | 3.910 | 1M | 51.0 | 29.0 | 38.8 |
| 13 | GPT-6.1 Sol (xhigh) | OpenAI | 闭源 | 51 | 0.390 | 1M | 64.0 | 69.2 | 77.0 |
| 14 | GPT-6 Astra (high) | OpenAI | 闭源 | 51 | 1.730 | 1M | 47.0 | 49.8 | 60.5 |
| 15 | GPT-6.1 Sol (high) | OpenAI | 闭源 | 50 | 0.320 | 1M | 65.0 | 58.0 | 65.7 |
| 16 | GPT-6 Astra (medium) | OpenAI | 闭源 | 50 | 1.540 | 1M | 45.0 | 5.9 | 16.9 |
| 17 | Claude Fable 5.1 (medium with fallback) | Anthropic | 闭源 | 49 | 2.980 | 1M | 50.0 | 8.4 | 18.3 |
| 18 | Muse Spark 1.3 (max) | Meta | 闭源 | 48 | 1.600 | 1M | 181.0 | 33.6 | 47.5 |
| 19 | GPT-6.1 Sol (medium) | OpenAI | 闭源 | 48 | 0.210 | 1M | 59.0 | 5.7 | 14.1 |
| 20 | GPT-6 Sol (max) | OpenAI | 闭源 | 48 | 1.050 | 872K | 76.0 | 185.7 | 192.2 |
| 21 | Claude Fable 5.1 (low with fallback) | Anthropic | 闭源 | 47 | 2.370 | 1M | 48.0 | 4.5 | 15.0 |
| 22 | Claude Sonnet 5.5 (high with fallback) | Anthropic | 闭源 | 47 | 1.080 | 1M | 92.0 | 14.5 | 19.9 |
| 23 | Grok 4.7 (xhigh) | SpaceXAI | 闭源 | 46 | 3.740 | 500K | 77.0 | 79.8 | 86.3 |
| 24 | Grok 4.7 (high) | SpaceXAI | 闭源 | 46 | 2.730 | 500K | 76.0 | 30.7 | 37.3 |
| 25 | MiMo-V2.6-Pro | Xiaomi | 开源 | 46 | 0.130 | 1M | 41.0 | 4.2 | 65.1 |
| 26 | GPT-6 Astra (low) | OpenAI | 闭源 | 46 | 0.820 | 1M | 46.0 | 3.0 | 13.9 |
| 27 | Qwen3.8 Max (0902) | Alibaba | 闭源 | 45 | 5.410 | 984K | 39.0 | 3.0 | 67.2 |
| 28 | Muse Spark 1.3 (xhigh) | Meta | 闭源 | 45 | 1.370 | 1M | 139.0 | 37.7 | 55.8 |
| 29 | GLM-5.3 (max) | Z AI | 开源 | 45 | 2.010 | 1M | 71.0 | 3.1 | 38.5 |
| 30 | GPT-6 Sol (xhigh) | OpenAI | 闭源 | 44 | 0.520 | 872K | 70.0 | 50.6 | 57.7 |
| 31 | Step 5 Preview | StepFun | 闭源 | 44 | 0.720 | 1M | 85.0 | 3.0 | 32.3 |
| 32 | Kimi K3 (max) | Kimi | 开源 | 44 | 2.000 | 1.05M | 36.0 | 4.0 | 72.9 |
| 33 | GPT-6 Sol (high) | OpenAI | 闭源 | 43 | 0.380 | 872K | 66.0 | 19.2 | 26.8 |
| 34 | Claude Opus 5.5 (low with fallback) | Anthropic | 闭源 | 42 | 0.550 | 1M | 72.0 | 14.2 | 21.2 |
| 35 | GPT-6.1 Sol (low) | OpenAI | 闭源 | 42 | 0.130 | 1M | 68.0 | 2.1 | 9.4 |
| 36 | GPT-5.6 Terra (max) | OpenAI | 闭源 | 42 | 1.400 | 1M | 93.0 | 173.8 | 179.2 |
| 37 | GLM-5.3-Flash | Z AI | 开源 | 42 | 0.250 | 1M | 45.0 | 3.4 | 59.2 |
| 38 | Gemini 3.8 Flash (high) | 闭源 | 41 | 1.240 | 1M | 242.0 | 20.8 | 22.8 | |
| 39 | Claude Sonnet 5.5 (medium with fallback) | Anthropic | 闭源 | 41 | 0.590 | 1M | 90.0 | 1.3 | 6.9 |
| 40 | Qwen3.8 2.4T A95B | Alibaba | 开源 | 40 | 2.160 | 262K | 39.0 | 2.9 | 66.6 |
| 41 | Qwen3.8-Flash-Next | Alibaba | 开源 | 40 | 0.370 | 256K | 58.0 | 2.5 | 45.4 |
| 42 | GPT-6 Sol (medium) | OpenAI | 闭源 | 40 | 0.250 | 872K | — | — | — |
| 43 | Gemini 3.8 Flash (medium) | 闭源 | 40 | 0.930 | 1M | — | — | — | |
| 44 | DeepSeek V4.1 Flash (max) | DeepSeek | 开源 | 39 | 0.270 | 1M | 209.0 | 0.9 | 12.9 |
| 45 | GPT-5.6 Terra (xhigh) | OpenAI | 闭源 | 38 | 0.630 | 1M | 84.0 | 28.9 | 34.9 |
| 46 | MiMo-V2.6-Flash | Xiaomi | 开源 | 38 | 0.060 | 1M | 57.0 | 4.2 | 48.4 |
| 47 | GPT-6 Luna (max) | OpenAI | 闭源 | 37 | 0.070 | 1M | 145.0 | 115.0 | 118.4 |
| 48 | DeepSeek V4 Pro 0813 (max) | DeepSeek | 开源 | 36 | 0.670 | 1M | 81.0 | 1.7 | 32.5 |
| 49 | DeepSeek V4 Flash Vision (max) | DeepSeek | 闭源 | 35 | 0.310 | 1M | 216.0 | 1.0 | 12.6 |
| 50 | GLM-5.3 (low) | Z AI | 开源 | 34 | 0.850 | 1M | 67.0 | 3.1 | 40.3 |
| 51 | GPT-5.6 Terra (high) | OpenAI | 闭源 | 34 | 0.340 | 1M | 79.0 | 2.6 | 8.9 |
| 52 | JT-4.1 Flash 236B A21B | China Mobile | 闭源 | 34 | — | 262K | — | — | — |
| 53 | GPT-6 Sol (low) | OpenAI | 闭源 | 34 | 0.130 | 872K | 68.0 | 2.3 | 9.6 |
| 54 | GPT-6 Luna (xhigh) | OpenAI | 闭源 | 34 | 0.040 | 1M | 131.0 | 20.1 | 23.9 |
| 55 | Qwen3.8 27B (xhigh) | Alibaba | 开源 | 34 | 1.010 | 256K | 44.0 | 3.9 | 61.0 |
| 56 | Motif 3 | Motif Technologies | 开源 | 34 | — | 262K | — | — | — |
| 57 | Gemini 3.8 Flash (low) | 闭源 | 33 | — | 1M | — | — | — | |
| 58 | GPT-5.3 Codex (xhigh) | OpenAI | 闭源 | 33 | — | 400K | 95.0 | 70.8 | 76.0 |
| 59 | GPT-6 Luna (high) | OpenAI | 闭源 | 32 | 0.030 | 1M | 124.0 | 15.2 | 19.2 |
| 60 | K2 Horizon 375B A23B | Institute of Foundation Models | 开源 | 31 | — | 524K | — | — | — |
| 61 | GPT-5.6 Terra (medium) | OpenAI | 闭源 | 30 | 0.180 | 1M | 77.0 | 1.9 | 8.4 |
| 62 | Kimi K3 (low) | Kimi | 开源 | 30 | 1.150 | 1.05M | 38.0 | 4.6 | 70.1 |
| 63 | Gemini 3.1 Pro Preview | 闭源 | 30 | 0.670 | 1M | 114.0 | 25.9 | 30.3 | |
| 64 | GPT-6 Luna (medium) | OpenAI | 闭源 | 29 | 0.020 | 1M | — | — | — |
| 65 | MiniMax-M3 | MiniMax | 开源 | 29 | 0.510 | 1M | 108.0 | 1.5 | 24.7 |
| 66 | Nex-N2-Pro | Nex AGI | 开源 | 28 | — | 262K | — | — | — |
| 67 | Solar Pro 4 | Upstage | 闭源 | 28 | — | 512K | 81.0 | 2.0 | 32.9 |
| 68 | GPT-6 Sol (Non-reasoning) | OpenAI | 闭源 | 28 | 0.330 | 872K | 69.0 | 1.0 | 8.3 |
| 69 | Qwen3.8 27B (medium) | Alibaba | 开源 | 28 | 1.130 | 256K | 48.0 | 3.9 | 55.9 |
| 70 | GPT-5.6 Terra (low) | OpenAI | 闭源 | 27 | 0.140 | 1M | 77.0 | 1.8 | 8.3 |
| 71 | JT-4.1 Flash 236B A21B (non-reasoning) | China Mobile | 闭源 | 27 | — | 256K | — | — | — |
| 72 | Quasar 438B (max) | Multiverse Computing | 闭源 | 27 | 2.020 | 1M | 134.0 | 1.2 | 19.8 |
| 73 | Apodex 1.1 | Apodex | 闭源 | 26 | 0.460 | 256K | — | — | — |
| 74 | Qwen3.8 27B (low) | Alibaba | 开源 | 26 | 1.050 | 256K | 49.0 | 3.9 | 54.5 |
| 75 | GPT-5.5 Instant (June 2026) | OpenAI | 闭源 | 26 | 0.690 | 400K | 111.0 | 1.0 | 23.6 |
| 76 | Kimi K2.7 Code | Kimi | 开源 | 26 | 0.540 | 256K | 75.0 | 2.8 | 39.2 |
| 77 | Inkling Small | Thinking Machines | 开源 | 26 | 0.090 | 1M | 200.0 | 2.0 | 14.5 |
| 78 | K2 Horizon MoVA 36B A4B | Institute of Foundation Models | 开源 | 25 | — | 524K | — | — | — |
| 79 | Hy3 | Tencent | 开源 | 25 | 0.070 | 256K | 87.0 | 3.0 | 31.8 |
| 80 | Qwen3.7 Plus | Alibaba | 闭源 | 25 | 0.220 | 1M | 57.0 | 2.2 | 46.3 |
| 81 | Inkling | Thinking Machines | 开源 | 25 | — | 1M | 168.0 | 1.8 | 16.7 |
| 82 | Solar Open2 250B | Upstage | 开源 | 25 | — | 1.05M | — | — | — |
| 83 | DeepSeek V4.1 Flash (Non-reasoning) | DeepSeek | 开源 | 25 | 0.150 | 1M | 207.0 | 1.1 | 3.5 |
| 84 | Ling-3.0-flash-VL | InclusionAI | 开源 | 25 | — | 262K | 142.0 | 1.9 | 19.4 |
| 85 | Nemotron 3 Ultra | NVIDIA | 开源 | 23 | 0.600 | 262K | 206.0 | 1.1 | 14.5 |
| 86 | A.X-K2 | SK Telecom | 开源 | 23 | — | 262K | — | — | — |
| 87 | Ling-3.0-flash-Fin | InclusionAI | 开源 | 23 | — | 262K | 261.0 | 1.8 | 11.4 |
| 88 | Gemini 3.5 Flash-Lite | 闭源 | 22 | 0.120 | 1M | 321.0 | 9.0 | 10.6 | |
| 89 | G9v3-39A5B | AI9Stars | 开源 | 22 | — | 131K | — | — | — |
| 90 | KAT-Coder-Pro V2 | KwaiKAT | 闭源 | 22 | — | 256K | — | — | — |
| 91 | Qwen3.5 397B A17B (Non-reasoning) | Alibaba | 开源 | 21 | — | 262K | 86.0 | 2.3 | 8.1 |
| 92 | GPT-6 Luna (low) | OpenAI | 闭源 | 21 | 0.004 | 1M | 124.0 | 2.0 | 6.0 |
| 93 | GPT-5.6 Terra (Non-reasoning) | OpenAI | 闭源 | 21 | 0.140 | 1M | 84.0 | 0.8 | 6.7 |
| 94 | K2 Horizon 7B | Institute of Foundation Models | 开源 | 21 | — | 524K | — | — | — |
| 95 | Qwen3.5 Omni Plus | Alibaba | 闭源 | 20 | — | 256K | 68.0 | 2.3 | 9.6 |
| 96 | DeepSeek V4 Pro 0813 (Non-reasoning) | DeepSeek | 开源 | 20 | 0.480 | 1M | 204.0 | 2.1 | 4.5 |
| 97 | o3 | OpenAI | 闭源 | 20 | — | 200K | 122.0 | 6.0 | 10.1 |
| 98 | Qwen3.8 27B | Alibaba | 开源 | 20 | 2.490 | 256K | 50.0 | 3.9 | 13.8 |
| 99 | Ling 3.0 Flash | InclusionAI | 开源 | 20 | — | 262K | 314.0 | 3.0 | 11.0 |
| 100 | K-EXAONE 2.0 | LG AI Research | 开源 | 20 | — | 262K | — | — | — |
| 101 | LongCat 2.0 | LongCat | 开源 | 19 | 0.060 | 1M | — | — | — |
| 102 | JT-35B-Flash | China Mobile | 闭源 | 19 | — | 256K | — | — | — |
| 103 | Qwen3.5 397B A17B | Alibaba | 开源 | 18 | 0.470 | 262K | 87.0 | 2.3 | 44.9 |
| 104 | GPT-6 Luna (Non-reasoning) | OpenAI | 闭源 | 18 | 0.010 | 1M | 139.0 | 0.8 | 4.4 |
| 105 | Qwen3.6 35B A3B | Alibaba | 开源 | 18 | 0.480 | 262K | 125.0 | 2.1 | 49.4 |
| 106 | Qwen3.5 122B A10B (Non-reasoning) | Alibaba | 开源 | 18 | — | 262K | 138.0 | 2.4 | 6.0 |
| 107 | Muse Glimmer (high) | Meta | 开源 | 17 | 0.060 | 131K | 145.0 | 1.3 | 18.6 |
| 108 | Doubao Seed Code | ByteDance Seed | 闭源 | 17 | — | 256K | — | — | — |
| 109 | Claude 4.5 Haiku | Anthropic | 闭源 | 17 | 0.280 | 200K | 91.0 | 16.0 | 21.5 |
| 110 | Gemma 4 26B A4B | 开源 | 17 | — | 256K | — | — | — | |
| 111 | Ring-2.6-1T | InclusionAI | 开源 | 17 | 0.290 | 262K | 116.0 | 4.2 | 25.8 |
| 112 | K2 Horizon 3.7B | Institute of Foundation Models | 开源 | 16 | — | 524K | — | — | — |
| 113 | Qwen3.5 122B A10B | Alibaba | 开源 | 16 | 0.320 | 262K | 125.0 | 2.4 | 22.5 |
| 114 | Claude 4.5 Haiku (Non-reasoning) | Anthropic | 闭源 | 15 | — | 200K | 83.0 | 0.7 | 6.7 |
| 115 | Qwen3.6 35B A3B (Non-reasoning) | Alibaba | 开源 | 15 | — | 262K | 139.0 | 2.2 | 5.8 |
| 116 | Granite 4.2 30B | IBM | 开源 | 15 | — | 131K | 77.0 | 0.9 | 33.3 |
| 117 | Gemma 4 31B | 开源 | 15 | 0.000 | 256K | 36.0 | 1.0 | 63.9 | |
| 118 | ERNIE 5.0 Thinking Preview | Baidu | 闭源 | 14 | — | 128K | — | — | — |
| 119 | Mistral Medium 3.5 | Mistral | 开源 | 14 | 0.500 | 256K | 165.0 | 2.3 | 17.4 |
| 120 | Gemma 4 12B | 开源 | 14 | — | 256K | 111.0 | 2.4 | 24.9 | |
| 121 | Nova 2.0 Pro Preview (medium) | Amazon | 闭源 | 14 | — | 256K | 113.0 | 16.0 | 38.0 |
| 122 | Gemma 4 31B (Non-reasoning) | 开源 | 14 | — | 256K | 38.0 | 2.0 | 15.0 | |
| 123 | Nova 2.0 Omni (medium) | Amazon | 闭源 | 14 | — | 1M | — | — | — |
| 124 | Apriel-v1.6-15B-Thinker | ServiceNow | 开源 | 13 | — | 128K | — | — | — |
| 125 | Nova 2.0 Lite (high) | Amazon | 闭源 | 13 | — | 1M | 188.0 | 12.5 | 25.8 |
| 126 | Qwen3.5 9B (Non-reasoning) | Alibaba | 开源 | 13 | — | 262K | — | — | — |
| 127 | EXAONE 4.5 33B | LG AI Research | 开源 | 13 | — | 262K | — | — | — |
| 128 | Command A+ | Cohere | 开源 | 13 | 0.000 | 192K | 209.0 | 0.5 | 12.4 |
| 129 | Gemma 4 26B A4B (Non-reasoning) | 开源 | 13 | — | 256K | 110.0 | 1.0 | 5.6 | |
| 130 | Qwen3.5 4B | Alibaba | 开源 | 13 | — | 262K | 29.0 | 0.8 | 86.5 |
| 131 | Nemotron 3.5 Lightning | NVIDIA | 开源 | 13 | 0.100 | 1M | 293.0 | 0.6 | 9.2 |
| 132 | Nemotron 3 Super | NVIDIA | 开源 | 13 | 1.640 | 1M | 164.0 | 1.3 | 16.6 |
| 133 | Nova 2.0 Pro Preview (low) | Amazon | 闭源 | 13 | — | 256K | 122.0 | 10.5 | 31.0 |
| 134 | Nova 2.0 Lite (medium) | Amazon | 闭源 | 12 | — | 1M | 177.0 | 15.0 | 29.1 |
| 135 | Qwen3.5 Omni Flash | Alibaba | 闭源 | 12 | — | 256K | 219.0 | 1.8 | 4.1 |
| 136 | MiniCPM5-2B | OpenBMB | 开源 | 12 | — | 131K | — | — | — |
| 137 | Mercury 2.5 | Inception | 闭源 | 12 | 0.120 | 260K | 637.0 | 3.5 | 4.3 |
| 138 | JT-MINI | China Mobile | 闭源 | 12 | — | 128K | — | — | — |
| 139 | Magistral Medium 1.2 | Mistral | 闭源 | 12 | — | 128K | — | — | — |
| 140 | Nova 2.0 Lite (low) | Amazon | 闭源 | 12 | — | 1M | 192.0 | 9.0 | 22.0 |
| 141 | HyperNova 60B 2605 (high) | Multiverse Computing | 开源 | 12 | — | 131K | — | — | — |
| 142 | Nemotron Cascade 2 30B A3B | NVIDIA | 开源 | 12 | — | 1M | — | — | — |
| 143 | gpt-oss-120b (high) | OpenAI | 开源 | 12 | 0.110 | 131K | 211.0 | 0.9 | 12.7 |
| 144 | K2 Think V2 | Institute of Foundation Models | 开源 | 11 | — | 262K | — | — | — |
| 145 | LongCat Flash Lite | LongCat | 开源 | 11 | — | 256K | — | — | — |
| 146 | HyperCLOVA X SEED Think (32B) | Naver | 开源 | 11 | — | 128K | — | — | — |
| 147 | Mistral Small 4 | Mistral | 开源 | 11 | 0.020 | 256K | 183.0 | 0.8 | 14.4 |
| 148 | Qwen3 Next 80B A3B (Reasoning) | Alibaba | 开源 | 11 | — | 262K | 169.0 | 2.4 | 17.1 |
| 149 | Qwen3.5 9B | Alibaba | 开源 | 11 | 0.210 | 262K | 36.0 | 2.4 | 72.1 |
| 150 | Nova 2.0 Omni (low) | Amazon | 闭源 | 11 | — | 1M | — | — | — |
| 151 | Granite 4.2 8B | IBM | 开源 | 11 | 0.020 | 131K | 62.0 | 0.8 | 40.9 |
| 152 | Ling 3.0 Tiny | InclusionAI | 开源 | 11 | 0.000 | 262K | 50.0 | 2.6 | 52.8 |
| 153 | Mi:dm K 2.5 Pro | Korea Telecom | 闭源 | 11 | — | 128K | — | — | — |
| 154 | G9v3-3B | AI9Stars | 开源 | 11 | — | 131K | — | — | — |
| 155 | Trinity Large Thinking | Arcee AI | 开源 | 11 | 0.120 | 512K | 336.0 | 1.3 | 8.7 |
| 156 | Qwen3.5 4B (Non-reasoning) | Alibaba | 开源 | 11 | — | 262K | 25.0 | 0.8 | 21.1 |
| 157 | INTELLECT-3 | Prime Intellect | 开源 | 11 | — | 131K | — | — | — |
| 158 | Solar Open 100B | Upstage | 开源 | 10 | — | 128K | — | — | — |
| 159 | Nemotron 3 Nano Omni 30B A3B | NVIDIA | 开源 | 10 | — | 256K | 262.0 | 0.4 | 9.9 |
| 160 | gpt-oss-120b (low) | OpenAI | 开源 | 10 | — | 131K | 211.0 | 0.9 | 12.7 |
| 161 | Llama 4 Maverick | Meta | 开源 | 10 | — | 1M | 96.0 | 0.9 | 6.2 |
| 162 | Nova 2.0 Pro Preview (Non-reasoning) | Amazon | 闭源 | 10 | — | 256K | 108.0 | 1.1 | 5.7 |
| 163 | gpt-oss-20b (low) | OpenAI | 开源 | 10 | — | 131K | 196.0 | 0.9 | 13.6 |
| 164 | North Mini Code | Cohere | 开源 | 10 | 0.000 | 256K | 68.0 | 0.4 | 37.2 |
| 165 | K2-V2 (high) | Institute of Foundation Models | 开源 | 10 | — | 512K | — | — | — |
| 166 | Qwen3 Next 80B A3B | Alibaba | 开源 | 10 | — | 262K | 170.0 | 2.2 | 5.1 |
| 167 | DiffusionGemma 26B A4B | 开源 | 10 | — | 256K | — | — | — | |
| 168 | Gemma 4 12B (Non-reasoning) | 开源 | 9 | — | 262K | 116.0 | 2.5 | 6.8 | |
| 169 | Mistral Large 3 | Mistral | 开源 | 9 | 0.030 | 256K | 78.0 | 1.1 | 7.5 |
| 170 | Qwen3 Coder Next | Alibaba | 开源 | 9 | 0.550 | 256K | 98.0 | 1.2 | 6.3 |
| 171 | Motif-2-12.7B | Motif Technologies | 闭源 | 9 | — | 128K | — | — | — |
| 172 | Nova Premier | Amazon | 闭源 | 9 | — | 1M | — | — | — |
| 173 | Granite 4.2 3B | IBM | 开源 | 9 | 0.010 | 131K | 224.0 | 0.5 | 11.7 |
| 174 | K2-V2 (medium) | Institute of Foundation Models | 开源 | 9 | — | 512K | — | — | — |
| 175 | Mistral Small 4 (Non-reasoning) | Mistral | 开源 | 9 | — | 256K | 165.0 | 0.8 | 3.8 |
| 176 | Tri-21B-Think | Trillion Labs | 开源 | 9 | — | 32K | — | — | — |
| 177 | gpt-oss-20b (high) | OpenAI | 开源 | 9 | 0.010 | 131K | 166.0 | 0.8 | 15.8 |
| 178 | Gemma 4 E4B | 开源 | 9 | — | 128K | 64.0 | 0.8 | 39.8 | |
| 179 | Nemotron 3 Nano | NVIDIA | 开源 | 9 | 0.020 | 1M | 193.0 | 1.1 | 14.1 |
| 180 | MiniCPM5-1B | OpenBMB | 开源 | 9 | — | 128K | — | — | — |
| 181 | Sarvam 105B (high) | Sarvam | 开源 | 9 | — | 128K | — | — | — |
| 182 | Nova 2.0 Lite (Non-reasoning) | Amazon | 闭源 | 9 | — | 1M | 174.0 | 1.0 | 3.9 |
| 183 | MiniCPM5-1B (Non-reasoning) | OpenBMB | 开源 | 9 | — | 128K | — | — | — |
| 184 | Magistral Small 1.2 | Mistral | 开源 | 9 | — | 128K | — | — | — |
| 185 | Nanbeige4.1-3B | Nanbeige | 开源 | 8 | — | 256K | — | — | — |
| 186 | LFM2.5-2.6B | Liquid AI | 开源 | 8 | — | 128K | — | — | — |
| 187 | EXAONE 4.0 32B | LG AI Research | 开源 | 8 | — | 131K | — | — | — |
| 188 | Nova 2.0 Omni (Non-reasoning) | Amazon | 闭源 | 8 | — | 1M | — | — | — |
| 189 | Llama 4 Scout | Meta | 开源 | 8 | — | 10M | 132.0 | 0.8 | 4.7 |
| 190 | Hermes 4 70B | Nous Research | 开源 | 8 | — | 128K | — | — | — |
| 191 | Falcon-H1R-7B | TII UAE | 开源 | 8 | — | 256K | — | — | — |
| 192 | Gemma 4 E2B | 开源 | 8 | — | 128K | — | — | — | |
| 193 | Qwen3 Omni 30B A3B (Reasoning) | Alibaba | 开源 | 8 | — | 65.5K | 103.0 | 1.9 | 26.3 |
| 194 | Step3 VL 10B | StepFun | 开源 | 8 | — | 65.5K | — | — | — |
| 195 | Llama 3.3 70B | Meta | 开源 | 8 | — | 128K | 87.0 | 1.6 | 7.4 |
| 196 | Llama Nemotron Ultra | NVIDIA | 开源 | 8 | — | 128K | — | — | — |
| 197 | ERNIE 4.5 300B A47B | Baidu | 开源 | 8 | — | 131K | — | — | — |
| 198 | Hermes 4 405B | Nous Research | 开源 | 7 | — | 128K | 42.0 | 2.4 | 62.4 |
| 199 | NVIDIA Nemotron Nano 12B v2 VL | NVIDIA | 开源 | 7 | — | 128K | — | — | — |
| 200 | Gemma 4 E4B (Non-reasoning) | 开源 | 7 | — | 128K | 66.0 | 0.8 | 8.3 | |
| 201 | NVIDIA Nemotron Nano 9B V2 | NVIDIA | 开源 | 7 | — | 131K | 104.0 | 5.9 | 30.0 |
| 202 | Hermes 4 405B (Non-reasoning) | Nous Research | 开源 | 7 | — | 128K | 41.0 | 2.4 | 14.5 |
| 203 | Nemotron 3 Nano 4B | NVIDIA | 开源 | 7 | — | 262K | — | — | — |
| 204 | K2-V2 (low) | Institute of Foundation Models | 开源 | 7 | — | 512K | — | — | — |
| 205 | Kimi Linear 48B A3B Instruct | Kimi | 开源 | 7 | — | 1M | — | — | — |
| 206 | Llama 3.1 405B | Meta | 开源 | 7 | — | 128K | — | — | — |
| 207 | LFM2.5-8B-A1B | Liquid AI | 开源 | 7 | — | 32.8K | — | — | — |
| 208 | Ring-flash-2.0 | InclusionAI | 开源 | 7 | — | 128K | — | — | — |
| 209 | Olmo 3.1 32B Think | Allen Institute for AI | 开源 | 7 | — | 65.5K | — | — | — |
| 210 | Command A | Cohere | 开源 | 7 | — | 256K | 60.0 | 1.7 | 10.1 |
| 211 | Qwen3.5 2B | Alibaba | 开源 | 7 | — | 262K | — | — | — |
| 212 | Nemotron 3 Nano (Non-reasoning) | NVIDIA | 开源 | 7 | — | 1M | 221.0 | 0.7 | 3.0 |
| 213 | NVIDIA Nemotron Nano 9B V2 (Non-reasoning) | NVIDIA | 开源 | 7 | — | 131K | 151.0 | 2.3 | 5.6 |
| 214 | Hermes 4 70B (Non-reasoning) | Nous Research | 开源 | 7 | — | 128K | — | — | — |
| 215 | Sarvam 30B (high) | Sarvam | 开源 | 7 | — | 65.5K | — | — | — |
| 216 | Olmo 3.1 32B Instruct | Allen Institute for AI | 开源 | 6 | — | 65.5K | — | — | — |
| 217 | Gemma 4 E2B (Non-reasoning) | 开源 | 6 | — | 128K | — | — | — | |
| 218 | R1 1776 | Perplexity | 开源 | 6 | — | 128K | — | — | — |
| 219 | Llama 3.2 90B (Vision) | Meta | 开源 | 6 | — | 128K | — | — | — |
| 220 | Celeris-1 | Celeris | 闭源 | 6 | 0.050 | 131K | 1,505.0 | 0.6 | 0.9 |
| 221 | Phi-4 Mini | Microsoft | 开源 | 6 | — | 128K | 45.0 | 0.8 | 11.9 |
| 222 | EXAONE 4.0 32B (Non-reasoning) | LG AI Research | 开源 | 6 | — | 131K | — | — | — |
| 223 | Qwen3.5 2B (Non-reasoning) | Alibaba | 开源 | 6 | — | 262K | — | — | — |
| 224 | Qwen3.5 0.8B | Alibaba | 开源 | 6 | — | 262K | — | — | — |
| 225 | DeepHermes 3 - Mistral 24B | Nous Research | 开源 | 6 | — | 32K | — | — | — |
| 226 | Jamba 1.7 Large | AI21 Labs | 开源 | 6 | — | 256K | — | — | — |
| 227 | Granite 4.0 H Small | IBM | 开源 | 6 | — | 128K | 14.0 | 35.2 | 70.1 |
| 228 | Ministral 3 14B | Mistral | 开源 | 6 | 0.020 | 256K | 77.0 | 1.0 | 7.5 |
| 229 | Qwen3 Omni 30B A3B | Alibaba | 开源 | 6 | — | 65.5K | 96.0 | 1.9 | 7.1 |
| 230 | LFM2 24B A2B | Liquid AI | 开源 | 6 | — | 32.8K | — | — | — |
| 231 | Phi-4 | Microsoft | 开源 | 6 | — | 16K | 40.0 | 2.6 | 15.1 |
| 232 | Nova Micro | Amazon | 闭源 | 6 | — | 130K | 252.0 | 0.9 | 2.9 |
| 233 | NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning) | NVIDIA | 开源 | 6 | — | 128K | 202.0 | 1.1 | 3.6 |
| 234 | Phi-4 Multimodal | Microsoft | 开源 | 6 | — | 128K | — | — | — |
| 235 | MiniCPM-V 4.6 1.3B | OpenBMB | 开源 | 6 | — | 262K | — | — | — |
| 236 | Jamba Reasoning 3B | AI21 Labs | 开源 | 6 | — | 262K | — | — | — |
| 237 | Reka Flash 3 | Reka AI | 开源 | 6 | — | 128K | — | — | — |
| 238 | Olmo 3 7B Think | Allen Institute for AI | 开源 | 6 | — | 65.5K | — | — | — |
| 239 | Molmo 7B-D | Allen Institute for AI | 开源 | 6 | — | 4.1K | — | — | — |
| 240 | Ministral 3 8B | Mistral | 开源 | 5 | 0.010 | 256K | 105.0 | 0.8 | 5.5 |
| 241 | Llama 3.2 11B (Vision) | Meta | 开源 | 5 | — | 128K | 14.0 | 2.5 | 38.9 |
| 242 | Qwen3.5 0.8B (Non-reasoning) | Alibaba | 开源 | 5 | — | 262K | — | — | — |
| 243 | Exaone 4.0 1.2B | LG AI Research | 开源 | 5 | — | 64K | — | — | — |
| 244 | Olmo 3 7B | Allen Institute for AI | 开源 | 5 | — | 65.5K | — | — | — |
| 245 | Exaone 4.0 1.2B (Non-reasoning) | LG AI Research | 开源 | 5 | — | 64K | — | — | — |
| 246 | LFM2.5-1.2B-Thinking | Liquid AI | 开源 | 5 | — | 32K | — | — | — |
| 247 | Jamba 1.7 Mini | AI21 Labs | 开源 | 5 | — | 258K | — | — | — |
| 248 | LFM2.5-1.2B-Instruct | Liquid AI | 开源 | 5 | — | 32K | — | — | — |
| 249 | Granite 4.0 H 1B | IBM | 开源 | 5 | — | 128K | — | — | — |
| 250 | Gemma 3 270M | 开源 | 5 | — | 32K | — | — | — | |
| 251 | Apertus 70B Instruct | Swiss AI Initiative | 开源 | 5 | — | 65.5K | — | — | — |
| 252 | Granite 4.0 Micro | IBM | 开源 | 5 | — | 128K | — | — | — |
| 253 | DeepHermes 3 - Llama-3.1 8B | Nous Research | 开源 | 5 | — | 128K | — | — | — |
| 254 | Molmo2-8B | Allen Institute for AI | 开源 | 5 | — | 36.9K | — | — | — |
| 255 | Ministral 3 3B | Mistral | 开源 | 5 | 0.010 | 256K | 201.0 | 0.8 | 3.3 |
| 256 | LFM2.5-VL-1.6B | Liquid AI | 开源 | 5 | — | 32K | — | — | — |
| 257 | Granite 4.0 350M | IBM | 开源 | 5 | — | 32.8K | — | — | — |
| 258 | Tiny Aya Global | Cohere | 开源 | 5 | — | 8.19K | — | — | — |
| 259 | Apertus 8B Instruct | Swiss AI Initiative | 开源 | 5 | — | 65.5K | — | — | — |
| 260 | Granite 4.0 H 350M | IBM | 开源 | 5 | — | 32.8K | — | — | — |
| 261 | K2 Horizon 0.9B | Institute of Foundation Models | 开源 | 3 | — | 131K | — | — | — |
相同厂商颜色相同，横向条形图按指数排序。
双子图共用横轴（任务平均成本，对数轴）：上图为全部模型散点 + 红色实线全局帕累托前沿；下图为各厂商自身的帕累托前沿（虚线，非前沿点已剔除）。右上角下拉选择器同时控制两个子图，坐标轴范围固定不变。⚠ 仅统计 AA 提供 API 定价的模型（108/261），免费($0)模型不适用对数成本轴未含在本图；开源自部署模型多无统一定价。