AI API Pricing by Provider

23 providers · 84 models · prices verified 2 October 2026

Every vendor's list pricing in one place: input and output rate per 1M tokens, context window size, and whether TokenCalc can count it exactly or only estimate. Numbers verified 2 October 2026.

Providers

01.AI (零一万物)1 models · Yi-Lightning (零一万物)
Alibaba Qwen (阿里通义)4 models · Qwen3 Max (通义千问), Qwen3 Plus, Qwen3 Flash, Qwen3 Coder
Amazon1 models · Amazon Nova Pro
Anthropic13 models · Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, Claude Fable 5 +8 more
Baichuan (百川智能)1 models · Baichuan 4 (百川)
Baidu ERNIE (百度)2 models · ERNIE 4.5 (文心一言), ERNIE X1
ByteDance Doubao (火山引擎)3 models · Doubao 1.6 Pro (豆包), Doubao 1.6 Flash, Doubao 1.6 Lite
DeepSeek (深度求索)4 models · DeepSeek V4 Pro, DeepSeek V4 Flash, DeepSeek V3.2, DeepSeek R1
Google8 models · Gemini 3.1 Pro, Gemini 3.6 Flash, Gemini 3.5 Flash, Gemini 3 Flash, Gemini 3.5 Flash-Lite +3 more
Huawei PanGu (华为云)1 models · PanGu 4.7 (盘古)
Kunlun (昆仑万维)1 models · SkyClaw 1.0 (昆仑万维)
Meta1 models · Llama 4 (Meta)
MiniMax (稀宇科技)1 models · MiniMax M2.5
Mistral AI1 models · Mistral Large 3
Moonshot Kimi (月之暗面)3 models · Kimi K2.6 (月之暗面), Kimi K2.5, Kimi K3
OpenAI24 models · GPT-5.5, GPT-5, GPT-5 mini, GPT-5 nano, GPT-4o +19 more
SenseTime (商汤)1 models · SenseNova 5.1 (商汤)
StepFun (阶跃星辰)1 models · Step-3 (阶跃星辰)
Tencent Hunyuan (腾讯)2 models · Hunyuan 2 Pro (混元), Hunyuan Lite
Xiaomi MiMo (小米)2 models · MiMo V2.5 Pro (小米), MiMo V2.5
Zhipu GLM (智谱)4 models · GLM-5 (智谱), GLM-5.1, GLM-4.7, GLM-4-Flash
iFlytek Spark (讯飞星火)2 models · Spark Max / X1 (讯飞星火), Spark Lite
xAI3 models · Grok 4.1 (xAI), Grok 4.5, Grok 4.3

What these numbers mean

  • List price — the public rate. Enterprise contracts, committed-use discounts and batch tiers are lower and are not reflected here.
  • Per 1M tokens — every vendor bills this way. Multiply by your token count divided by a million.
  • Output costs more — on most models the output rate is 3–8× the input rate, so long generations dominate the bill.
  • Context is a hard ceiling — exceeding it fails the request; it does not bill over. Check it before you optimise price.

The same data is available as JSON under CC BY 4.0.