LLM API 定价对比 (2026)
上次验证 2026-05-23 · 价格单位为每百万 token（美元）· 除特别说明外使用标准 API 档位
对比主流前沿模型和预算模型的 API 输入、输出成本。每一行都链接到用于核验的官方定价页面；如果你在比较订阅制 AI 编码助手，请查看 coding plan 定价对比。
相关对比： coding plan 定价 · 图像 API 定价 · 所有对比 English
| 提供商 | 模型 | 输入 $/MTok | 输出 $/MTok | 上下文 | 来源 |
|---|---|---|---|---|---|
| 💰 预算档（输入低于 $1/MTok） | |||||
| OpenAI | GPT-5 nano | $0.05 | $0.4 | 400K | openai.com |
| OpenAI | GPT-4.1 nano | $0.1 | $0.4 | 1M | openai.com |
| Gemini 2.0 Flash | $0.1 | $0.4 | 1M | ai.google.dev | |
| Gemini 2.5 Flash Lite | $0.1 | $0.4 | 1M | ai.google.dev | |
| DeepSeek | Deepseek V4 Flash | $0.14 | $0.28 | 1M | api-docs.deepseek.com |
| OpenAI | GPT-4o mini | $0.15 | $0.6 | 128K | openai.com |
| OpenAI | GPT-5.4 nano | $0.2 | $1.25 | 400K | openai.com |
| OpenAI | GPT-5 mini | $0.25 | $2 | 400K | openai.com |
| Gemini 3.1 Flash Lite | $0.25 | $1.5 | 1M | ai.google.dev | |
| Gemini 2.5 Flash | $0.3 | $2.5 | 1M | ai.google.dev | |
| OpenAI | GPT-4.1 mini | $0.4 | $1.6 | 1M | openai.com |
| DeepSeek | Deepseek V4 Pro | $0.43 | $0.87 | 1M | api-docs.deepseek.com |
| OpenAI | GPT-5.4 mini | $0.75 | $4.5 | 400K | openai.com |
| 📊 中档（输入 $1–3/MTok） | |||||
| Anthropic | Claude Haiku 4.5 | $1 | $5 | 1M | claude.com |
| OpenAI | o4-mini | $1.1 | $4.4 | 200K | openai.com |
| OpenAI | GPT-5 | $1.25 | $10 | 1M | openai.com |
| OpenAI | GPT-5.1 | $1.25 | $10 | 1M | openai.com |
| Gemini 2.5 Pro | $1.25 | $10 | 1M | ai.google.dev | |
| Gemini 3.5 Flash | $1.5 | $9 | 1M | ai.google.dev | |
| OpenAI | GPT-5.2 | $1.75 | $14 | 1M | openai.com |
| OpenAI | GPT-4.1 | $2 | $8 | 1M | openai.com |
| OpenAI | o3 | $2 | $8 | 200K | openai.com |
| OpenAI | GPT-4o | $2.5 | $10 | 128K | openai.com |
| OpenAI | GPT-5.4 | $2.5 | $15 | 1M | openai.com |
| ⭐ 高端档（输入 $3+/MTok） | |||||
| Anthropic | Claude Sonnet 4 | $3 | $15 | 1M | claude.com |
| Anthropic | Claude Sonnet 4.5 | $3 | $15 | 1M | claude.com |
| Anthropic | Claude Sonnet 4.6 | $3 | $15 | 1M | claude.com |
| Anthropic | Claude Opus 4.5 | $5 | $25 | 1M | claude.com |
| Anthropic | Claude Opus 4.6 | $5 | $25 | 1M | claude.com |
| Anthropic | Claude Opus 4.7 | $5 | $25 | 1M | claude.com |
| OpenAI | GPT-5.5 | $5 | $30 | 1M | openai.com |
| Anthropic | Claude Opus 4 | $15 | $75 | 1M | claude.com |
| Anthropic | Claude Opus 4.1 | $15 | $75 | 1M | claude.com |
| OpenAI | GPT-5 pro | $15 | $120 | 1M | openai.com |
| OpenAI | GPT-5.4 pro | $30 | $180 | 1M | openai.com |
| OpenAI | GPT-5.5 pro | $30 | $180 | 1M | openai.com |
| 🔓 开源权重模型（经 API 托管商提供，非厂商标价） | |||||
| Meta | Llama 4 Maverick (hosted) | $0.05–0.90 | $0.05–0.90 | 1M | groq.com · docs.fireworks.ai |
| DeepSeek | DeepSeek R1 (hosted) | See official pricing | See official pricing | 128K | api-docs.deepseek.com |
| Alibaba | Qwen3 Max (DashScope) | See official pricing | See official pricing | 128K | help.aliyun.com |
* 区间价格取决于托管方或具体 SKU；请查看链接中的官方页面。除特别说明外，OpenAI 行使用 Standard 档。
📊 输出成本可视化对比 ($/MTok)
💡 关键发现
下方分析内容来自英文真源，第一版暂不做动态翻译。
Output often costs more than input
Most providers charge higher $/MTok for output than input. Shorter prompts and concise output formats reduce monthly spend.
Budget models cover many workloads
Flash / mini / Haiku-class models are sufficient for many coding and content tasks; reserve Opus / GPT-5.4 pro for the hardest reasoning.
Open-weight hosting varies
Llama and similar models are priced by the inference host (Groq, Fireworks, etc.), not a single vendor list price—always check your provider's page.
Promotions change quickly
DeepSeek and others run time-limited discounts. Re-check official pages before locking budgets; this table shows last_verified date in the header.
🧮 快速成本估算器
预估月费用
—
📋 官方定价页面
- OpenAI API Pricing — openai.com
- Anthropic API Pricing — claude.com
- Google Gemini API Pricing — ai.google.dev
- DeepSeek API Pricing — api-docs.deepseek.com
- Groq Pricing — groq.com
- Fireworks AI Serverless Pricing — docs.fireworks.ai
- Alibaba Model Studio (Qwen) Pricing — help.aliyun.com
以上价格为上次验证日期时的官方标价。服务商可能随时调整价格；预算前请务必打开官方页面再次确认。