python-genai: Structured outputs is not working for gemini-2.5-pro-preview-03-25
Steps to reproduce
Just follow the example in the docs at https://ai.google.dev/gemini-api/docs/structured-output?lang=python.
If we do:
import os
from google import genai
from pydantic import BaseModel
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
client = genai.Client(api_key=GEMINI_API_KEY)
MODEL = "gemini-2.0-flash"
class Recipe(BaseModel):
recipe_name: str
ingredients: list[str]
response = client.models.generate_content(
model=MODEL,
contents="List a few popular cookie recipes. Be sure to include the amounts of ingredients.",
config={
"response_mime_type": "application/json",
"response_schema": list[Recipe],
},
)
print(response.text)
my_recipes: list[Recipe] = response.parsed
print(my_recipes)
Everything works and we get the following output:
[
{
"recipe_name": "Chocolate Chip Cookies",
"ingredients": [
"1 cup (2 sticks) unsalted butter, softened",
"3/4 cup granulated sugar",
"3/4 cup packed brown sugar",
"1 teaspoon vanilla extract",
"2 large eggs",
"2 1/4 cups all-purpose flour",
"1 teaspoon baking soda",
"1 teaspoon salt",
"2 cups chocolate chips"
]
},
{
"recipe_name": "Peanut Butter Cookies",
"ingredients": [
"1 cup (2 sticks) unsalted butter, softened",
"1 cup granulated sugar",
"1 cup packed brown sugar",
"1 teaspoon vanilla extract",
"2 large eggs",
"1 cup peanut butter",
"2 1/2 cups all-purpose flour",
"1 teaspoon baking soda",
"1/2 teaspoon salt"
]
},
{
"recipe_name": "Sugar Cookies",
"ingredients": [
"1 1/2 cups (3 sticks) unsalted butter, softened",
"2 cups granulated sugar",
"4 large eggs",
"1 teaspoon vanilla extract",
"5 cups all-purpose flour",
"2 teaspoons baking powder",
"1 teaspoon salt"
]
}
]
[Recipe(recipe_name='Chocolate Chip Cookies', ingredients=['1 cup (2 sticks) unsalted butter, softened', '3/4 cup granulated sugar', '3/4 cup packed brown sugar', '1 teaspoon vanilla extract', '2 large eggs', '2 1/4 cups all-purpose flour', '1 teaspoon baking soda', '1 teaspoon salt', '2 cups chocolate chips']), Recipe(recipe_name='Peanut Butter Cookies', ingredients=['1 cup (2 sticks) unsalted butter, softened', '1 cup granulated sugar', '1 cup packed brown sugar', '1 teaspoon vanilla extract', '2 large eggs', '1 cup peanut butter', '2 1/2 cups all-purpose flour', '1 teaspoon baking soda', '1/2 teaspoon salt']), Recipe(recipe_name='Sugar Cookies', ingredients=['1 1/2 cups (3 sticks) unsalted butter, softened', '2 cups granulated sugar', '4 large eggs', '1 teaspoon vanilla extract', '5 cups all-purpose flour', '2 teaspoons baking powder', '1 teaspoon salt'])]
If we change the model to gemini-2.5-pro-preview-03-25 or gemini-2.5-pro-exp-03-25 and keep everything else identical:
import os
from google import genai
from pydantic import BaseModel
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
client = genai.Client(api_key=GEMINI_API_KEY)
MODEL = "gemini-2.5-pro-preview-03-25"
class Recipe(BaseModel):
recipe_name: str
ingredients: list[str]
response = client.models.generate_content(
model=MODEL,
contents="List a few popular cookie recipes. Be sure to include the amounts of ingredients.",
config={
"response_mime_type": "application/json",
"response_schema": list[Recipe],
},
)
print(response.text)
my_recipes: list[Recipe] = response.parsed
print(my_recipes)
We’ll get an output like:
ny
```json
[
{
"recipe_name": "Classic Chocolate Chip Cookies",
"ingredients": [
"1 cup (2 sticks) unsalted butter, softened",
"3/4 cup granulated sugar",
"3/4 cup packed light brown sugar",
"2 large eggs",
"1 teaspoon vanilla extract",
"2 1/4 cups all-purpose flour",
"1 teaspoon baking soda",
"1/2 teaspoon salt",
"2 cups semi-sweet chocolate chips"
]
},
{
"recipe_name": "Simple Peanut Butter Cookies",
"ingredients": [
"1 cup creamy peanut butter",
"1 cup granulated sugar",
"1 large egg",
"1 teaspoon vanilla extract (optional)",
"1/2 teaspoon baking soda"
]
},
{
"recipe_name": "Oatmeal Raisin Cookies",
"ingredients": [
"1 cup (2 sticks) unsalted butter, softened",
"1 cup packed light brown sugar",
"1/2 cup granulated sugar",
"2 large eggs",
"1 teaspoon vanilla extract",
"1 1/2 cups all-purpose flour",
"1 teaspoon baking soda",
"1 teaspoon ground cinnamon",
"1/2 teaspoon salt",
"3 cups old-fashioned rolled oats",
"1 cup raisins"
]
},
{
"recipe_name": "Sugar Cookies",
"ingredients": [
"2 3/4 cups all-purpose flour",
"1 teaspoon baking soda",
"1/2 teaspoon salt",
"1 cup (2 sticks) unsalted butter, softened",
"1 1/2 cups granulated sugar",
"1 large egg",
"1 teaspoon vanilla extract"
]
}
]
``
None
Something is broken for Gemini 2.5 structured outputs. response.text includes markdown ticks among other things, and response.parsed is None.
About this issue
- Original URL
- State: closed
- Created 2 years ago
- Reactions: 11
- Comments: 30
@pamorgan can we please bump up the priority of this one to a P1? This impacts basically all API consumers of Google AI Studio.
Thank you for reporting. We are investigating the issue.
Fix has been rolled out to Gemini API - Please let us know if you are still seeing the issue.
@benpml we don’t typically share ETAs externally. will share once it’s done 😃
We have identified the root cause and working on a fix soon
Same thing for us:
Thank you - I’m closing this now. Please let us know if you continue to see the issue.
Yes - We will continue updating the issue.
Will you update us when it’s pushed to the VertexAI API? Still seeing it with VertexAI.