统计 Gemini 模型的词元数和计费字符数

生成式模型会将数据细分为单元（称为“词元”）以进行处理。每个模型具有在提示和响应中可以处理的词元数上限。

本页介绍了如何使用 Count Tokens API 估算 对 Gemini 模型的请求的词元数和计费字符数。没有用于获取响应中令牌估算值的 API。

请注意，Count Tokens API 不能用于 Imagen 模型。

计数中会提供哪些信息？

请注意以下关于计算词元数和计费字符数的说明：

  • 统计总令牌数

    • 此计数有助于确保您的请求不会超出允许的上下文窗口。

    • 令牌数将反映在请求输入中提供的所有文件（例如图片）的大小。系统不会统计视频中的图片数量或秒数。

    • 对于所有 Gemini 模型，一个令牌大约相当于 4 个字符。100 个词元大约相当于 60-80 个英语单词。

  • 统计计费字符总数

    • 此计数有助于了解和控制费用，因为对于 Vertex AI，字符数是价格计算的一部分。

    • 可结算字符数将反映在请求输入中提供的文本中的字符数。

对于较旧的 Gemini 型号，令牌不包含在价格计算中；但对于 Gemini 2.0 型号，令牌会用于价格计算。详细了解每个模型的令牌限制和每个模型的价格。

计算词元和计费字符数的价格和配额

使用 CountTokens API 无需付费或配额限制。CountTokens API 的最大配额为每分钟 3000 个请求 (RPM)。

代码示例

仅限文本输入

Swift

let response = try await model.countTokens("Write a story about a magic backpack.")
print("Total Tokens: \(response.totalTokens)")
print("Total Billable Characters: \(response.totalBillableCharacters)")

Kotlin 

val response = generativeModel.countTokens("Write a story about a magic backpack.")
println("Total Tokens: ${response.totalTokens}")
println("Total Billable Characters: ${response.totalBillableCharacters}")

Java 

Content prompt = new Content.Builder()
    .addText("Write a story about a magic backpack.")
    .build();

GenerativeModelFutures modelFutures = GenerativeModelFutures.from(model);
ListenableFuture<CountTokensResponse> countTokensResponse =
    modelFutures.countTokens(prompt);

Futures.addCallback(countTokensResponse, new FutureCallback<CountTokensResponse>() {
    @Override
    public void onSuccess(CountTokensResponse response) {
        System.out.println("Total Tokens = " + response.getTotalTokens());
        System.out.println("Total Billable Characters: = " +
          response.getTotalBillableCharacters());
    }

    @Override
    public void onFailure(Throwable t) {
        t.printStackTrace();
    }
}, executor);

Web 

const { totalTokens, totalBillableCharacters } = await model.countTokens("Write a story about a magic backpack.");
console.log(`Total tokens: ${totalTokens}, total billable characters: ${totalBillableCharacters}`);

Dart 

final tokenCount = await model.countTokens(Content.text("Write a story about a magic backpack."));
print('Token count: ${tokenCount.totalTokens}, billable characters: ${tokenCount.totalBillableCharacters}');

多模态输入

Swift

let response = try await model.countTokens(image, "What's in this picture?")
print("Total Tokens: \(response.totalTokens)")
print("Total Billable Characters: \(response.totalBillableCharacters)")

Kotlin 

val prompt = content {
  image(bitmap)
  text("What's in this picture?")
}
val response = generativeModel.countTokens(prompt)
println("Total Tokens: ${response.totalTokens}")
println("Total Billable Characters: ${response.totalBillableCharacters}")

Java 

Content prompt = new Content.Builder()
    .addImage(bitmap)
    .addText("What's in this picture?")
    .build();

GenerativeModelFutures modelFutures = GenerativeModelFutures.from(model);
ListenableFuture<CountTokensResponse> countTokensResponse =
    modelFutures.countTokens(prompt);

Futures.addCallback(countTokensResponse, new FutureCallback<CountTokensResponse>() {
    @Override
    public void onSuccess(CountTokensResponse response) {
        System.out.println("Total Tokens = " + response.getTotalTokens());
        System.out.println("Total Billable Characters: = " +
          response.getTotalBillableCharacters());
    }

    @Override
    public void onFailure(Throwable t) {
        t.printStackTrace();
    }
}, executor);

Web 

const prompt = "What's in this picture?";
const imagePart = { inlineData: { mimeType: 'image/jpeg', data: imageAsBase64 }};

const { totalTokens, totalBillableCharacters } = await model.countTokens([prompt, imagePart]);
console.log(`Total tokens: ${totalTokens}, total billable characters: ${totalBillableCharacters}`);

Dart 

final prompt = TextPart("What's in the picture?");
final tokenCount = await model.countTokens([
  Content.multi([prompt, imagePart])
]);
print('Token count: ${tokenCount.totalTokens}, billable characters: ${tokenCount.totalBillableCharacters}');