Consistantly seeing longer response time for gemini-2.5-flash-preview-05-20

[florianmartens]

For me, simple object generation queries tend to be ~30% slower with the Vercel AI SDK vs. Google's native GoogleGenAI library.

I'm comparing these two code blocks:

    const ai = new GoogleGenAI({ apiKey: this.apiKey });

    const result = await ai.models.generateContent({
      model: "gemini-2.5-flash-preview-05-20",
      contents: promptText,
      config: {
        responseMimeType: "application/json",
        maxOutputTokens: 500,
        temperature: 0.2,
        thinkingConfig: {
          thinkingBudget: 0,
        },
        responseSchema: {
          type: Type.OBJECT,
          properties: {
            name: {
              type: Type.STRING,
              nullable: true,
            },
            linkedinUrl: {
              type: Type.STRING,
              nullable: true,
            },
            websiteUrl: {
              type: Type.STRING,
              nullable: true,
            },
          },
        },
      },
    });

   const { object } = await generateObject({
      model: google("gemini-2.5-flash-preview-04-17"),
      providerOptions: {
        thinkingConfig: { thinkingBudget: 0, includeThoughts: false },
      } satisfies GoogleGenerativeAIProviderOptions,
      schema: IdentifiersSchema,
      temperature: 0.2,
      maxTokens: 500,
      prompt: promptText,
    });

Measuring response times is baked into my app. All the 7-10s response times use the vercel AI SDk. The 5-6s response times use the native GoogleGenAI SDK.

is this to be expected. Are people seeing similar things? What could be the cause of this?

[lgrammel]

what library versions are you using?

[florianmartens]

Hey @lgrammel ! I'm using @ai-sdk/google": 1.2.19 and ai: 4.3.16.

[lgrammel]

can you try this with ai sdk 5? we have swapped out our stream processing so it should be faster