综合总榜
共 28 个模型。默认按 v5 智能指数排序，未收录 v5 的模型排在其后并以 v4.3.2 分数显示灰色值。点击任意列头可切换排序。
| # | 模型 / 厂商 | v5 | v4.3.2 | GPQA | TB 4.0 | LBench D | 输入 | 输出 | tok/s | 上下文 | 类型 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 CarbonGoogle | 59.4 | — | — | 65.2 | — | $2.00 | $10.00 | 87 | 1M | 闭源 |
| 2 | Claude Opus 5.5Anthropic | 58.4 | 57.6 | 93.9 | 65.2 | 86.9 | $4.00 | $20.00 | 96 | 1M | 闭源 |
| 3 | GPT-6 AstraOpenAI | 56.9 | — | 95.8 | — | 82.0 | $10.00 | $50.00 | 81 | 1.05M | 闭源 |
| 4 | GPT-6.1 SolOpenAI | 54.5 | — | 95.4 | — | 82.1 | $2.00 | $10.00 | 77 | 1.05M | 闭源 |
| 5 | Claude Sonnet 5.5Anthropic | 54.2 | 56.0 | 95.6 | 64.0 | 85.0 | $2.00 | $10.00 | 135 | 1M | 闭源 |
| 6 | GPT-6.1 TerraOpenAI | 51.0 | — | 94.3 | 55.6 | 82.3 | $1.00 | $5.00 | 94 | 1.05M | 闭源 |
| 7 | Gemini 4 ArgonGoogle | 50.4 | — | 96.4 | — | 73.7 | $2.00 | $10.00 | 80 | 1M | 闭源 |
| 8 | Claude Fable 5.1Anthropic | 49.7 | 53.0 | — | — | 83.4 | $10.00 | $50.00 | 57 | 1M | 闭源 |
| 9 | Kimi K3.1 Pro 开源Moonshot AI | 46.0 | 55.1 | — | — | 77.8 | $2.00 | $10.00 | 28 | 1M | 开源 |
| 10 | Muse WonderMeta | 43.3 | — | — | — | — | $2.00 | $10.00 | 143 | 1M | 闭源 |
| 11 | Claude Haiku 5.5Anthropic | 36.2 | 43.0 | — | — | 83.1 | $0.10 | $0.50 | 235 | 1M | 闭源 |
| 12 | Grok 4.7xAI | 35.0 | 46.4 | 92.7 | — | 78.0 | $2.00 | $6.00 | 105 | 500K | 闭源 |
| 13 | DeepSeek V4.1 Flash 开源DeepSeek | 32.5 | 45.1 | 89.8 | 19.7 | 83.4 | $0.30 | $1.20 | 148 | 1.05M | 开源 |
| 14 | GLM-5.3 开源Z.ai | 32.0 | 44.8 | 90.9 | 38.9 | — | $0.60 | $2.20 | 77 | 1M | 开源 |
| 15 | MiMo-V2.6 Pro 开源Xiaomi | 30.0 | 46.3 | 59.5 | 31.3 | 78.5 | $0.35 | $1.40 | 81 | 1M | 开源 |
| 16 | MiniMax M3.1 FlashMiniMax | 29.0 | — | — | — | 74.0 | $0.15 | $0.60 | 118 | 1M | 闭源 |
| 17 | DeepSeek V4 Pro 开源DeepSeek | 23.0 | — | — | — | — | $1.32 | $3.96 | 72 | 1.05M | 开源 |
| 18 | Kimi K3 开源Moonshot AI | — | 43.6 | 93.1 | 17.2 | 77.2 | $2.00 | $15.00 | 30 | 1M | 开源 |
| 19 | GLM-5.3 Flash 开源Z.ai | — | 41.8 | — | — | 77.2 | $0.20 | $0.80 | 120 | 1M | 开源 |
| 20 | Qwen3.8 27B 开源Alibaba Qwen | — | 33.7 | — | — | 75.9 | $0.10 | $0.40 | 180 | 262K | 开源 |
| 21 | MiniMax M3 开源MiniMax | — | 29.2 | 90.9 | 1.0 | — | $0.25 | $1.00 | 126 | 1M | 开源 |
| 22 | Gemma 4 31B 开源Google | — | 19.0 | 75.8 | — | 58.4 | $0.10 | $0.40 | 200 | 262K | 开源 |
| 23 | GPT-6 LunaOpenAI | — | — | — | — | 82.5 | $0.10 | $0.50 | 130 | 1.05M | 闭源 |
| 24 | Gemini 3.8 FlashGoogle | — | — | 95.4 | — | 72.7 | $0.75 | $3.75 | 160 | 1.05M | 闭源 |
| 25 | Qwen3.8 MaxAlibaba Qwen | — | — | 92.7 | 34.3 | 78.6 | $2.00 | $8.00 | 88 | 1M | 闭源 |
| 26 | gpt-oss-120b 开源OpenAI | — | — | 75.8 | — | — | $0.05 | $0.20 | 220 | 131K | 开源 |
| 27 | Step 5 PreviewStepFun | — | — | — | — | 76.9 | $0.50 | $2.00 | 95 | 1M | 闭源 |
| 28 | Nemotron 3 Ultra 开源NVIDIA | — | — | — | — | 62.8 | $0.40 | $1.60 | 70 | 1M | 开源 |
为什么有两列智能指数
AA Intelligence Index 于 2026-10-27 从 v4.3.2 升到 v5，评分口径变了。同一个 DeepSeek V4.1 Flash 在旧版是 45.1、新版是 32.5。两组数字必须分开看，混排会得出完全错误的结论。
“—”不是 0 分
表示该基准没有收录此模型的公开结果，而不是它得了零分。开源模型和刚发布的模型更容易缺数据，用综合分横向比时要留意样本偏差。
价格是标准档
均为各厂商第一方 API 的官方标价，美元 / 百万 tokens，不含批量折扣、不含企业协议价。部分模型有促销价，到期时间见价格页。