为什么按一个答案算价，而不是按一百万 token？
因为没人按一百万 token 买。服务商对输入和输出分别按不同费率报价，所以两个模型按百万看着接近，在同一件真实任务上却可能差很多。按一份固定工作量算价，数字才有可比性。
同样一个答案，换个模型价格能相差 2,000 倍。
服务商公布的是每百万 token 的价格，这在模型之间无法比较。本指数把一个标准答案，1,000 个输入 token 加 500 个输出 token，在来自 29 家提供商的 100 个模型上算出价格。最后更新于 2026年8月20日。
关键数字
- 最便宜的答案：Ling-3.0-flash（inclusionAI）上 $0.000053。
- 中位数答案：Kimi K2 Thinking 上 $0.0019。
- 最贵的答案：Claude Opus 4.7 (Fast)（Anthropic）上 $0.105。
- 从最便宜到最贵的差距：约 2,000 倍。
计算方式
一个标准答案是 1,000 个输入 token，大致相当于带几段上下文的提示词，加上 500 个输出 token，大致相当于一段充实的回复。每个模型都按这同一份工作量计价：（输入费率 x 1,000 + 输出费率 x 500）除以 1,000,000。
改变配比，排序就会变，这正是重点：输入便宜、输出贵的模型在短提示上占优，在长回复上吃亏。公式已公开，你可以按自己的配比重新计算。
来源与局限
- 价格是 OpenRouter 模型目录中的公开标价，按每百万 token 的美元计，于 2026年8月20日 获取。
- 不含提示词缓存折扣、批量定价或协商的用量费率。规模化之后的实际账单会更低。
- 成本不等于质量。便宜 2,000 倍的模型并不会差 2,000 倍，在许多日常任务上差距很小。
- 积分一列是 Whizi 在自家套餐中按消息收取的量，只对单独计量的 89 个模型显示。它不是提供商的费率。
完整指数
100 个模型，从便宜到贵。标注 Whizi AI Model Cost Index 并链接到本页即可自由再利用。原始数据：JSON 和 CSV。
| # | 模型 | 提供商 | 每个答案 | 每 1,000 个答案 | $ / 百万输入 | $ / 百万输出 | 上下文 | Whizi 积分 |
|---|---|---|---|---|---|---|---|---|
| 1 | Ling-3.0-flash | inclusionAI | $0.000053 | $0.052 | $0.021 | $0.063 | 262K | 1x |
| 2 | Nex-N2-Mini | Nex Agi | $0.000075 | $0.075 | $0.025 | $0.1 | 262K | 1x |
| 3 | Solar Pro 4 | Upstage | $0.000090 | $0.090 | $0.03 | $0.12 | 524K | 1x |
| 4 | Qwen3.7 Flash | Qwen | $0.000095 | $0.095 | $0.03 | $0.13 | 1M | 1x |
| 5 | Granite 4.1 8B | Ibm Granite | $0.000100 | $0.100 | $0.05 | $0.1 | 131K | 1x |
| 6 | Laguna XS 2.1 | Poolside | $0.000120 | $0.120 | $0.06 | $0.12 | 262K | 1x |
| 7 | Hy-MT2-1.8B | Tencent | $0.000133 | $0.133 | $0.044 | $0.177 | 8K | 不计量 |
| 8 | Phi 4 | Microsoft | $0.000140 | $0.140 | $0.07 | $0.14 | 16K | 1x |
| 9 | Nemotron 3.5 Lightning | NVIDIA | $0.000180 | $0.180 | $0.08 | $0.2 | 262K | 1x |
| 10 | Laguna S 2.1 | Poolside | $0.000180 | $0.180 | $0.09 | $0.18 | 1M | 1x |
| 11 | Hy-MT2-30B-A3B | Tencent | $0.000222 | $0.222 | $0.074 | $0.295 | 8K | 不计量 |
| 12 | Llama 3.3 70B Instruct | Meta | $0.000260 | $0.260 | $0.1 | $0.32 | 131K | 1x |
| 13 | DeepSeek V4 Flash 0731 | DeepSeek | $0.000280 | $0.280 | $0.14 | $0.28 | 1M | 1x |
| 14 | Gemini 2.5 Flash Lite | $0.000300 | $0.300 | $0.1 | $0.4 | 1M | 1x | |
| 15 | Ring-2.6-1T | inclusionAI | $0.000388 | $0.388 | $0.075 | $0.625 | 262K | 1x |
| 16 | Hy3 | Tencent | $0.000396 | $0.396 | $0.132 | $0.528 | 262K | 1x |
| 17 | KAT-Coder-Air V2.5 | Kwaipilot | $0.000450 | $0.450 | $0.15 | $0.6 | 256K | 1x |
| 18 | DeepSeek V3.2 | DeepSeek | $0.000469 | $0.469 | $0.269 | $0.4 | 164K | 1x |
| 19 | Qwen3 Coder Next | Qwen | $0.000520 | $0.520 | $0.12 | $0.8 | 262K | 1x |
| 20 | Llama 4 Maverick | Meta | $0.000600 | $0.600 | $0.2 | $0.8 | 1M | 1x |
| 21 | Qwen3.6 35B A3B | Qwen | $0.000640 | $0.640 | $0.14 | $1 | 262K | 1x |
| 22 | DeepSeek V3.1 | DeepSeek | $0.000725 | $0.725 | $0.25 | $0.95 | 164K | 1x |
| 23 | Codestral 2508 | Mistral | $0.000750 | $0.750 | $0.3 | $0.9 | 256K | 1x |
| 24 | Qwen3.6 Flash | Qwen | $0.000750 | $0.750 | $0.1875 | $1.125 | 1M | 1x |
| 25 | Nex-N2-Pro | Nex Agi | $0.000750 | $0.750 | $0.25 | $1 | 262K | 1x |
| 26 | MiniMax M2 | MiniMax | $0.000765 | $0.765 | $0.255 | $1.02 | 205K | 1x |
| 27 | Step 3.7 Flash | Stepfun | $0.000775 | $0.775 | $0.2 | $1.15 | 262K | 1x |
| 28 | GPT-5.6 Luna Pro | OpenAI | $0.000800 | $0.800 | $0.2 | $1.2 | 1M | 不计量 |
| 29 | GPT-5.6 Luna | OpenAI | $0.000800 | $0.800 | $0.2 | $1.2 | 1M | 1x |
| 30 | GPT-5.4 Nano | OpenAI | $0.000825 | $0.825 | $0.2 | $1.25 | 400K | 2x |
| 31 | MiniMax M3 | MiniMax | $0.000900 | $0.900 | $0.3 | $1.2 | 1M | 1x |
| 32 | LongCat 2.0 | Meituan | $0.000900 | $0.900 | $0.3 | $1.2 | 1M | 1x |
| 33 | Perceptron Mk1 | Perceptron | $0.000900 | $0.900 | $0.15 | $1.5 | 33K | 1x |
| 34 | Qwen3.7 Plus | Qwen | $0.000960 | $0.960 | $0.32 | $1.28 | 1M | 1x |
| 35 | Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) | $0.0010 | $1.000 | $0.25 | $1.5 | 66K | 不计量 | |
| 36 | Gemini 3.1 Flash Lite | $0.0010 | $1.000 | $0.25 | $1.5 | 1M | 1x | |
| 37 | Inkling Small | Thinkingmachines | $0.0010 | $1.050 | $0.45 | $1.2 | 524K | 1x |
| 38 | Muse Glimmer 30B | Meta | $0.0011 | $1.100 | $0.35 | $1.5 | 131K | 2x |
| 39 | Qwen3.5 Plus 2026-04-20 | Qwen | $0.0012 | $1.200 | $0.3 | $1.8 | 1M | 1x |
| 40 | Mistral Large 3 2512 | Mistral | $0.0013 | $1.250 | $0.5 | $1.5 | 262K | 2x |
| 41 | GLM 4.7 | Z.ai | $0.0013 | $1.275 | $0.4 | $1.75 | 205K | 2x |
| 42 | Gemini 3.7 Flash | $0.0013 | $1.313 | $0.375 | $1.875 | 1M | 2x | |
| 43 | Mistral Medium 3.1 | Mistral | $0.0014 | $1.400 | $0.4 | $2 | 131K | 2x |
| 44 | Aion-3.0-Mini | Aion Labs | $0.0014 | $1.400 | $0.7 | $1.4 | 131K | 2x |
| 45 | GLM 4.6 | Z.ai | $0.0015 | $1.500 | $0.5 | $2 | 205K | 1x |
| 46 | Gemini 2.5 Flash | $0.0015 | $1.550 | $0.3 | $2.5 | 1M | 2x | |
| 47 | Gemini 3.5 Flash Lite | $0.0015 | $1.550 | $0.3 | $2.5 | 1M | 2x | |
| 48 | GLM 5 | Z.ai | $0.0016 | $1.560 | $0.6 | $1.92 | 205K | 2x |
| 49 | Kimi K2 0711 | Moonshot | $0.0017 | $1.720 | $0.57 | $2.3 | 131K | 10x |
| 50 | Seed 2.1 Turbo | Bytedance Seed | $0.0018 | $1.750 | $0.5 | $2.5 | 262K | 2x |
| 51 | Kimi K2 Thinking | Moonshot | $0.0019 | $1.850 | $0.6 | $2.5 | 262K | 10x |
| 52 | R1 | DeepSeek | $0.0019 | $1.950 | $0.7 | $2.5 | 64K | 2x |
| 53 | Seed-2.0-Code | Bytedance Seed | $0.0020 | $2.000 | $0.5 | $3 | 262K | 3x |
| 54 | Nano Banana 2 (Gemini 3.1 Flash Image) | $0.0020 | $2.000 | $0.5 | $3 | 131K | 不计量 | |
| 55 | Grok Build 0.1 | xAI | $0.0020 | $2.000 | $1 | $2 | 256K | 3x |
| 56 | Qwen3.8 27B | Qwen | $0.0021 | $2.050 | $0.45 | $3.2 | 1M | 3x |
| 57 | KAT-Coder-Pro V2.5 | Kwaipilot | $0.0022 | $2.220 | $0.74 | $2.96 | 256K | 3x |
| 58 | Nova Pro 1.0 | Amazon | $0.0024 | $2.400 | $0.8 | $3.2 | 300K | 3x |
| 59 | Nemotron 3 Ultra | NVIDIA | $0.0024 | $2.400 | $0.6 | $3.6 | 512K | 3x |
| 60 | Kimi K2.7 Code | Moonshot | $0.0025 | $2.460 | $0.71 | $3.5 | 262K | 10x |
| 61 | GLM 5.2 | Z.ai | $0.0025 | $2.484 | $0.966 | $3.036 | 1M | 3x |
| 62 | Grok 4.3 | xAI | $0.0025 | $2.500 | $1.25 | $2.5 | 1M | 4x |
| 63 | Gemini 3.6 Flash | $0.0026 | $2.625 | $0.75 | $3.75 | 1M | 3x | |
| 64 | Sakana Namazu | Sakana | $0.0029 | $2.950 | $0.95 | $4 | 262K | 4x |
| 65 | DeepSeek V4 Pro 0813 | DeepSeek | $0.0030 | $2.970 | $1.188 | $3.564 | 1M | 2x |
| 66 | Inkling | Thinkingmachines | $0.0030 | $2.975 | $0.95 | $4.05 | 1M | 4x |
| 67 | GPT-5.4 Mini | OpenAI | $0.0030 | $3.000 | $0.75 | $4.5 | 400K | 4x |
| 68 | Muse Spark 1.2 | Meta | $0.0034 | $3.375 | $1.25 | $4.25 | 1M | 5x |
| 69 | Muse Spark 1.1 | Meta | $0.0034 | $3.375 | $1.25 | $4.25 | 1M | 5x |
| 70 | Claude Haiku 4.5 | Anthropic | $0.0035 | $3.500 | $1 | $5 | 200K | 4x |
| 71 | GLM 5.3 | Z.ai | $0.0036 | $3.600 | $1.4 | $4.4 | 1M | 5x |
| 72 | Qwen3.7 Max | Qwen | $0.0037 | $3.688 | $1.475 | $4.425 | 1M | 5x |
| 73 | Grok 4.5 | xAI | $0.0050 | $5.000 | $2 | $6 | 500K | 6x |
| 74 | Qwen3.8 Max | Qwen | $0.0050 | $5.000 | $2 | $6 | 1M | 6x |
| 75 | Qwen3.8 2.4T A95B | Qwen | $0.0050 | $5.000 | $2 | $6 | 1M | 8x |
| 76 | Grok 4.6 | xAI | $0.0050 | $5.000 | $2 | $6 | 500K | 6x |
| 77 | Mistral Medium 3.5 | Mistral | $0.0053 | $5.250 | $1.5 | $7.5 | 262K | 6x |
| 78 | Gemini 3.5 Flash | $0.0060 | $6.000 | $1.5 | $9 | 1M | 8x | |
| 79 | Aion-3.0 | Aion Labs | $0.0060 | $6.000 | $3 | $6 | 131K | 10x |
| 80 | Claude Sonnet 5 | Anthropic | $0.0070 | $7.000 | $2 | $10 | 1M | 10x |
| 81 | Command A | Cohere | $0.0075 | $7.500 | $2.5 | $10 | 256K | 10x |
| 82 | Gemini 3.1 Pro Preview | $0.0080 | $8.000 | $2 | $12 | 1M | 10x | |
| 83 | GPT-5.6 Terra Pro | OpenAI | $0.0080 | $8.000 | $2 | $12 | 1M | 不计量 |
| 84 | GPT-5.6 Terra | OpenAI | $0.0080 | $8.000 | $2 | $12 | 1M | 4x |
| 85 | Nano Banana Pro (Gemini 3 Pro Image) | $0.0080 | $8.000 | $2 | $12 | 131K | 不计量 | |
| 86 | GPT-5.4 | OpenAI | $0.010 | $10.000 | $2.5 | $15 | 1M | 15x |
| 87 | GPT-5.6 Sol Pro | OpenAI | $0.010 | $10.000 | $2.5 | $15 | 1M | 不计量 |
| 88 | GPT-5.6 Sol | OpenAI | $0.010 | $10.000 | $2.5 | $15 | 1M | 20x |
| 89 | Claude Sonnet 4.6 | Anthropic | $0.011 | $10.500 | $3 | $15 | 1M | 10x |
| 90 | Claude Sonnet 4.5 | Anthropic | $0.011 | $10.500 | $3 | $15 | 1M | 10x |
| 91 | Kimi K3 | Moonshot | $0.011 | $10.500 | $3 | $15 | 1M | 10x |
| 92 | Claude Opus 5 | Anthropic | $0.018 | $17.500 | $5 | $25 | 1M | 20x |
| 93 | Claude Opus 4.8 | Anthropic | $0.018 | $17.500 | $5 | $25 | 1M | 20x |
| 94 | GPT-5.5 | OpenAI | $0.020 | $20.000 | $5 | $30 | 1M | 20x |
| 95 | GPT Chat Latest | OpenAI | $0.020 | $20.000 | $5 | $30 | 400K | 25x |
| 96 | Fugu Ultra | Sakana | $0.020 | $20.000 | $5 | $30 | 1M | 25x |
| 97 | Claude Opus 5 (Fast) | Anthropic | $0.035 | $35.000 | $10 | $50 | 1M | 不计量 |
| 98 | Claude Fable 5 | Anthropic | $0.035 | $35.000 | $10 | $50 | 1M | 50x |
| 99 | Claude Opus 4.8 (Fast) | Anthropic | $0.035 | $35.000 | $10 | $50 | 1M | 不计量 |
| 100 | Claude Opus 4.7 (Fast) | Anthropic | $0.105 | $105.000 | $30 | $150 | 1M | 不计量 |
常见问题
哪个模型最划算？
这要看任务。有用的结论是，这张表的中段每个答案只相差几百分之一美分，所以对大多数日常工作来说，好模型之间的价格差可以忽略，选择应该看输出质量。
多久更新一次？
价格会从服务商目录重新拉取，页面上标有抓取日期，目前是 2026年8月20日。这个市场的费率经常变动，引用数字前请先看日期。
Whizi 积分是什么？
Whizi 按消息从每月额度中扣减，贵的模型比便宜的消耗更多积分，从 1 倍到 20 倍不等。这一列显示的就是该倍数，方便你看出服务商的标价如何对应到一个统一套餐。
如何引用
一份数据集被再利用的次数，取决于它是否容易标注来源，这是可以直接复制的一行：
Whizi AI Model Cost Index, prices as of 2026-08-20. https://whizi.io/zh/tools/model-cost-index/
以 CC BY 4.0 发布，只要写明指数名称并链接到本页，即可用于商业用途。发表前需要核实数字的作者可通过 [email protected] 联系维护者。
下一步
用 token 费用计算器 为自己的文本算价，或在 Whizi 价格页 比较套餐。