OpenAI compatibility

Source: https://ai.google.dev/gemini-api/docs/openai#javascript

Skip to main con­tent

Mod­els → https://ai.google.dev/gemini-api/docs

Gem­ini API docs → https://ai.google.dev/gemini-api/docs

API Ref­er­ence → https://ai.google.dev/api
Cook­book → https://github.com/google-gemini/cookbook
Com­mu­nity → https://discuss.ai.google.dev/c/gemini-api/
Solu­tions
Code assis­tance
Show­case
Com­mu­nity
Get started
Overview → https://ai.google.dev/gemini-api/docs
Quick­start → https://ai.google.dev/gemini-api/docs/quickstart
API keys → https://ai.google.dev/gemini-api/docs/api-key
Libraries → https://ai.google.dev/gemini-api/docs/libraries
Ope­nAI com­pat­ib ­ il­ity → https://ai.google.dev/gemini-api/docs/openai
Mod­els
All mod­els → https://ai.google.dev/gemini-api/docs/models
Pric­ing → https://ai.google.dev/gemini-api/docs/pricing
Rate lim­its → https://ai.google.dev/gemini-api/docs/rate-limits
Billing info → https://ai.google.dev/gemini-api/docs/billing
Model Capa­bil­it­ ies
Text gen­er­a­tion → https://ai.google.dev/gemini-api/docs/text-generation
Image gen­er­a­tion → https://ai.google.dev/gemini-api/docs/image-generati
on
Video gen­er­a­tion → https://ai.google.dev/gemini-api/docs/video

1
Speech gen­er­a­tion → https://ai.google.dev/gemini-api/docs/speech-genera
tion
Music gen­er­a­tion → https://ai.google.dev/gemini-api/docs/music-generatio
n
Long con­text → https://ai.google.dev/gemini-api/docs/long-context
Struc­tured out­put → https://ai.google.dev/gemini-api/docs/structured-out
put
Think­ing → https://ai.google.dev/gemini-api/docs/thinking
Func­tion call­ing → https://ai.google.dev/gemini-api/docs/function-calling
Doc­u­ment under­stand­ing → https://ai.google.dev/gemini-api/docs/docu
ment-processing
Image under­stand­ing → https://ai.google.dev/gemini-api/docs/image-und
erstanding
Video under­stand­ing → https://ai.google.dev/gemini-api/docs/video-unde
rstanding
Audio under­stand­ing → https://ai.google.dev/gemini-api/docs/audio
Code exe­cu­tion → https://ai.google.dev/gemini-api/docs/code-execution
URL con­text → https://ai.google.dev/gemini-api/docs/url-context
Google Search → https://ai.google.dev/gemini-api/docs/google-search
Guides
Prompt engi­neer­ing → https://ai.google.dev/gemini-api/docs/prompting-s
trategies

Con­text caching → https://ai.google.dev/gemini-api/docs/caching

Files API → https://ai.google.dev/gemini-api/docs/files
Token count­ing → https://ai.google.dev/gemini-api/docs/tokens
Embed­dings → https://ai.google.dev/gemini-api/docs/embeddings

Resources
Migrate to Gen AI SDK → https://ai.google.dev/gemini-api/docs/migrate
Release notes → https://ai.google.dev/gemini-api/docs/changelog

2
 API trou­bleshoot­ing → https://ai.google.dev/gemini-api/docs/troubleshoo
ting
Fine-​tuning → https://ai.google.dev/gemini-api/docs/model-tuning

Poli­cies
Terms of ser­vice → https://ai.google.dev/gemini-api/terms
Avail­able regions → https://ai.google.dev/gemini-api/docs/available-region
s
Addi­tional usage polices → https://ai.google.dev/gemini-api/docs/usage-
policies
Gem­ini mod­els are acces­si­ble using the Ope­nAI libraries (Python
and Type­Script / Javascript) along with the REST API, by updat­ing
three lines of code and using your Gem­ini API key → https://aistudio.go
ogle.com/apikey. If you aren't already using the Ope­nAI libraries, we
rec­om­mend that you call the Gem­ini API directly → https://ai.google.de
v/gemini-api/docs/quickstart.

3
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

response = client.chat.com­ple­tions.cre­
ate(
model="gemini-​2.5-​
flash",
mes­
sages=[
{"role": "sys­
tem", "con­tent": "You are a help­
ful
assis­
tant."},
{
"role": "user",
"con­
tent": "Explain to me how AI works"
}
]
)

print(response.choices[0].mes­
sage)

4
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

const response = await ope­nai.chat.com­

ple­
tions.cre­ate({
model: "gemini-​2.0-​
flash",
mes­
sages: [
{ role: "sys­
tem", con­
tent: "You are a help­ful
assis­
tant." },
{
role: "user",
con­
tent: "Explain to me how AI works",
},
],
});

con­
sole.log(response.choices[0].mes­
sage);

5
REST

curl "https://gen­er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/chat/com­
ple­
tions" \
-H "Content-​Type: appli­
ca­tion/json" \
-H "Autho­riza­
tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "gemini-​2.0-​
flash",
"mes­
sages": [
{"role": "user", "con­ tent": "Explain to me how AI
works"}
]
}'

What changed? Just three lines!

INI_API_KEY": Replace "GEM­
api_key="GEM­ INI_API_KEY" with your
actual Gem­ini API key, which you can get in Google AI Stu­dio
→ https://aistudio.google.com/.
base_url="https://gen­
er­
a­
tive­
lan­
‐
nai/": This tells the Ope­nAI
guage.googleapis.com/v1beta/ope­
library to send requests to the Gem­ini API end­point instead of
the default URL.
model="gemini-​ flash": Choose a com­pat­ib
2.0-​ ­ le Gem­ini model

Think­ing
Gem­ini 2.5 mod­els are trained to think through com­plex prob­lems,
lead­ing to sig­nif­i­cantly improved rea­son­ing. The Gem­ini API comes
with a "think­ing bud­get" para­me­ter → https://ai.google.dev/gemini-api/doc
s/thinking#set-budget which gives fine grain con­trol over how much the
model will think.

6
 Unlike the Gem­ini API, the Ope­nAI API offers three lev­els of
think­ing con­trol: "low", "medium", and "high", which map to 1,024,
8,192, and 24,576 tokens, respec­tively.
If you want to dis­able think­ing, you can set rea­son­ing_effort to
"none" (note that rea­son­ing can­not be turned off for 2.5 Pro mod­els).

Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

response = client.chat.com­ple­tions.cre­
ate(
model="gemini-​2.5-​
flash",
rea­
son­ing_effort="low",
mes­
sages=[
{"role": "sys­
tem", "con­tent": "You are a help­
ful
assis­
tant."},
{
"role": "user",
"con­
tent": "Explain to me how AI works"
}
]
)

print(response.choices[0].mes­
sage)

7
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

const response = await ope­nai.chat.com­

ple­
tions.cre­ate({
model: "gemini-​2.5-​
flash",
rea­
son­ing_effort: "low",
mes­
sages: [
{ role: "sys­
tem", con­
tent: "You are a help­ful
assis­
tant." },
{
role: "user",
con­
tent: "Explain to me how AI works",
},
],
});

con­
sole.log(response.choices[0].mes­
sage);

8
REST

curl "https://gen­er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/chat/com­
ple­
tions" \
-H "Content-​Type: appli­
ca­tion/json" \
-H "Autho­riza­
tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "gemini-​2.5-​
flash",
"rea­
son­ing_effort": "low",
"mes­
sages": [
{"role": "user", "con­ tent": "Explain to me how AI
works"}
]
}'

Gem­ini think­ing mod­els also pro­duce thought sum­maries → https://ai.

google.dev/gemini-api/docs/thinking#summaries and can use exact think­ing
bud­gets → https://ai.google.dev/gemini-api/docs/thinking#set-budget. You can
use the extra_body field to include these fields in your request.
Note that rea­son­ing_effort and think­ing_bud­get over­lap func­‐
tion­al­ity, so they can't be used at the same time.

9
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

response = client.chat.com­ ple­

tions.cre­
ate(
model="gemini-​2.5-​
flash",
mes­
sages=[{"role": "user", "con­ tent": "Explain to me
how AI works"}],
extra_body={
'extra_body': {
"google": {
"think­
ing_con­
fig": {
"think­
ing_bud­
get": 800,
"include_thoughts": True
}
}
}
}
)

print(response.choices[0].mes­
sage)

10
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

const response = await ope­nai.chat.com­ple­

tions.cre­
ate({
model: "gemini-​2.5-​
flash",
mes­
sages: [{role: "user", con­ tent: "Explain to me how
AI works",}],
extra_body: {
"google": {
"think­
ing_con­
fig": {
"think­
ing_bud­
get": 800,
"include_thoughts": true
}
}
}
});

con­
sole.log(response.choices[0].mes­
sage);

11
REST

curl "https://gen­ er­

a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/chat/com­
ple­
tions" \
-H "Content-​ Type: appli­
ca­tion/json" \
-H "Autho­riza­tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "gemini-​ 2.5-​
flash",
"mes­sages": [{"role": "user", "con­ tent": "Explain to
me how AI works"}],
"extra_body": {
"google": {
"think­ing_con­
fig": {
"include_thoughts": true
}
}
}
}'

Stream­ing
The Gem­ini API sup­ports stream­ing responses → https://ai.google.dev/ge
mini-api/docs/text-generation?lang=python#generate-a-text-stream.

12
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

response = client.chat.com­ple­tions.cre­
ate(
model="gemini-​
2.0-​flash",
mes­
sages=[
{"role": "sys­tem", "con­tent": "You are a help­
ful
assis­
tant."},
{"role": "user", "con­tent": "Hello!"}
],
stream=True
)

for chunk in response:

print(chunk.choices[0].delta)

13
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

async func­tion main() {

const com­ple­
tion = await ope­nai.chat.com­
ple­
tions.cre­‐
ate({
model: "gemini-​2.0-​
flash",
mes­sages: [
{"role": "sys­
tem", "con­
tent": "You are a help­ful
assis­
tant."},
{"role": "user", "con­
tent": "Hello!"}
],
stream: true,
});

for await (const chunk of com­

ple­
tion) {
con­
sole.log(chunk.choices[0].delta.con­
tent);
}
}

main();

14
REST

curl "https://gen­er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/chat/com­
ple­
tions" \
-H "Content-​Type: appli­
ca­tion/json" \
-H "Autho­riza­
tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "gemini-​2.0-​
flash",
"mes­
sages": [
{"role": "user", "con­ tent": "Explain to me how AI
works"}
],
"stream": true
}'

Func­tion call­ing
Func­tion call­ing makes it eas­ier for you to get struc­tured data out­‐
puts from gen­er­a­tive mod­els and is sup­ported in the Gem­ini API → h
ttps://ai.google.dev/gemini-api/docs/function-calling/tutorial.

15
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

tools = [
{
"type": "func­ tion",
"func­tion": {
"name": "get_weather",
"descrip­tion": "Get the weather in a given loca­
‐
tion",
"para­
me­ters": {
"type": "object",
"prop­er­
ties": {
"loca­tion": {
"type": "string",
"descrip­
tion": "The city and state, e.g.
Chicago, IL",
},
"unit": {"type": "string", "enum": ["cel­sius",
"fahren­heit"]},
},
"required": ["loca­tion"],
},
}
}
]

mes­
sages = [{"role": "user", "con­
tent": "What's the

16
weather like in Chicago today?"}]
response = client.chat.com­ple­
tions.cre­
ate(
model="gemini-​
2.0-​flash",
mes­
sages=mes­
sages,
tools=tools,
tool_choice="auto"
)

print(response)

17
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

async func­tion main() {

const mes­sages = [{"role": "user", "con­ tent": "What's
the weather like in Chicago today?"}];
const tools = [
{
"type": "func­ tion",
"func­tion": {
"name": "get_weather",
"descrip­ tion": "Get the weather in a given loca­
‐
tion",
"para­ me­
ters": {
"type": "object",
"prop­er­
ties": {
"loca­tion": {
"type": "string",
"descrip­
tion": "The city and state, e.g.
Chicago, IL",
},
"unit": {"type": "string", "enum": ["cel­‐
sius", "fahren­ heit"]},
},
"required": ["loca­tion"],
},
}
}

18
 ];

const response = await ope­

nai.chat.com­
ple­
tions.cre­
ate({
model: "gemini-​
2.0-​
flash",
mes­
sages: mes­
sages,
tools: tools,
tool_choice: "auto",
});

con­
sole.log(response);
}

main();

19
REST

curl "https://gen­er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­ nai/chat/com­
ple­
tions" \
-H "Content-​Type: appli­ca­
tion/json" \
-H "Autho­
riza­tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "gemini-​2.0-​
flash",
"mes­
sages": [
{
"role": "user",
"con­
tent": "What'\''s the weather like in Chicago
today?"
}
],
"tools": [
{
"type": "func­tion",
"func­
tion": {
"name": "get_weather",
"descrip­tion": "Get the cur­rent weather in a given
loca­
tion",
"para­me­
ters": {
"type": "object",
"prop­er­
ties": {
"loca­tion": {
"type": "string",
"descrip­
tion": "The city and state, e.g.
Chicago, IL"
},
"unit": {
"type": "string",
"enum": ["cel­sius", "fahren­
heit"]
}
},

20
 "required": ["loca­
tion"]
}
}
}
],
"tool_choice": "auto"
}'

Image under­stand­ing
Gem­ini mod­els are natively mul­ti­modal and pro­vide best in class
per­for­mance on many com­mon vision tasks → https://ai.google.dev/gemi
ni-api/docs/vision.

21
Python

import base64
from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

# Func­
tion to encode the image
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return
base64.b64encode(image_file.read()).decode('utf-8')

# Get­
ting the base64 string
base64_image = encode_image("Path/to/agi/image.jpeg")

response = client.chat.com­ ple­

tions.cre­
ate(
model="gemini-​ 2.0-​
flash",
mes­sages=[
{
"role": "user",
"con­tent": [
{
"type": "text",
"text": "What is in this image?",
},
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,
{base64_image}"

22
 },
},
],
}
],
)

print(response.choices[0])

23
JavaScript

import Ope­
nAI from "ope­
nai";
import fs from 'fs/promises';

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

async func­
tion encodeIm­age(imagePath) {
try {
const image­
Buffer = await fs.read­File(imagePath);
return image­Buffer.toString('base64');
} catch (error) {
con­
sole.error("Error encod­ing image:", error);
return null;
}
}

async func­
tion main() {
const imagePath = "Path/to/agi/image.jpeg";
const base64Image = await encodeIm­
age(imagePath);

const mes­sages = [
{
"role": "user",
"con­tent": [
{
"type": "text",
"text": "What is in this image?",
},
{
"type": "image_url",

24
 "image_url": {
"url": `data:image/jpeg;base64,${base64Image}`
},
},
],
}
];

try {
const response = await ope­
nai.chat.com­
ple­
tions.cre­
‐
ate({
model: "gemini-​
2.0-​
flash",
mes­
sages: mes­
sages,
});

con­
sole.log(response.choices[0]);
} catch (error) {
con­
sole.error("Error call­
ing Gem­
ini API:", error);
}
}

main();

25
REST

bash -c '
base64_image=$(base64 -i "Path/to/agi/image.jpeg");
curl "https://gen­
er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­nai/chat/com­
ple­
tions" \
-H "Content-​
Type: appli­
ca­
tion/json" \
-H "Autho­
riza­
tion: Bearer GEM­INI_API_KEY" \
-d "{
\"model\": \"gemini-​
2.0-​flash\",
\"mes­
sages\": [
{
\"role\": \"user\",
\"con­
tent\": [
{ \"type\": \"text\", \"text\": \"What is in
this image?\" },
{
\"type\": \"image_url\",
\"image_url\": { \"url\":
\"data:image/jpeg;base64,${base64_image}\" }
}
]
}
]
}"
'

Gen­er­ate an image
Gen­er­ate an image:

26
Python

import base64
from ope­
nai import Ope­nAI
from PIL import Image
from io import Byte­
sIO

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/",
)

response = client.images.gen­er­
ate(
model="imagen-​3.0-​
generate-002",
prompt="a por­trait of a sheepadoo­
dle wear­
ing a cape",
response_for­
mat='b64_json',
n=1,
)

for image_data in response.data:

image = Image.open(Byte­
‐
sIO(base64.b64decode(image_data.b64_json)))
image.show()

27
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/",
});

async func­tion main() {

const image = await ope­nai.images.gen­er­
ate(
{
model: "imagen-​
3.0-​
generate-002",
prompt: "a por­
trait of a sheepadoo­dle wear­
ing a
cape",
response_for­
mat: "b64_json",
n: 1,
}
);

con­
sole.log(image.data);
}

main();

28
REST

curl "https://gen­er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/images/gen­
er­
a­
tions" \
-H "Content-​
Type: appli­ ca­
tion/json" \
-H "Autho­
riza­tion: Bearer GEM­INI_API_KEY" \
-d '{
"model": "imagen-​3.0-​generate-002",
"prompt": "a por­trait of a sheepadoo­dle wear­
ing a
cape",
"response_for­mat": "b64_json",
"n": 1,
}'

Audio under­stand­ing
Ana­lyze audio input:

29
Python

import base64
from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

with open("/path/to/your/audio/file.wav", "rb") as

audio_file:
base64_audio =
base64.b64encode(audio_file.read()).decode('utf-8')

response = client.chat.com­ ple­

tions.cre­
ate(
model="gemini-​ 2.0-​
flash",
mes­
sages=[
{
"role": "user",
"con­ tent": [
{
"type": "text",
"text": "Tran­ scribe this audio",
},
{
"type": "input_audio",
"input_audio": {
"data": base64_audio,
"for­
mat": "wav"
}
}
],
}

30
 ],
)

print(response.choices[0].mes­
sage.con­
tent)

31
JavaScript

import fs from "fs";

import Ope­
nAI from "ope­
nai";

const client = new Ope­

nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/",
});

const audioFile = fs.read­

File­
‐
Sync("/path/to/your/audio/file.wav");
const base64Audio =
Buffer.from(audioFile).toString("base64");

async func­ tion main() {

const response = await client.chat.com­ ple­
tions.cre­
ate({
model: "gemini-​ 2.0-​
flash",
mes­
sages: [
{
role: "user",
con­tent: [
{
type: "text",
text: "Tran­
scribe this audio",
},
{
type: "input_audio",
input_audio: {
data: base64Audio,
for­
mat: "wav",
},
},
],

32
 },
],
});

con­
sole.log(response.choices[0].mes­
sage.con­
tent);
}

main();

33
REST

bash -c '
base64_audio=$(base64 -i
"/path/to/your/audio/file.wav");
curl "https://gen­
er­
a­
tive­
lan­‐
guage.googleapis.com/v1beta/ope­nai/chat/com­
ple­
tions" \
-H "Content-​
Type: appli­
ca­
tion/json" \
-H "Autho­
riza­
tion: Bearer GEM­INI_API_KEY" \
-d "{
\"model\": \"gemini-​
2.0-​flash\",
\"mes­
sages\": [
{
\"role\": \"user\",
\"con­
tent\": [
{ \"type\": \"text\", \"text\": \"Tran­scribe
this audio file.\" },
{
\"type\": \"input_audio\",
\"input_audio\": {
\"data\": \"${base64_audio}\",
\"for­
mat\": \"wav\"
}
}
]
}
]
}"
'

34
Struc­tured out­put
Gem­ini mod­els can out­put JSON objects in any struc­ture you define
→ https://ai.google.dev/gemini-api/docs/structured-output.

Python

from pydan­
tic import Base­
Model
from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

class Cal­
en­darEvent(Base­
Model):
name: str
date: str
par­
tic­
i­
pants: list[str]

com­
ple­tion = client.beta.chat.com­ ple­
tions.parse(
model="gemini-​2.0-​flash",
mes­
sages=[
{"role": "sys­
tem", "con­ tent": "Extract the event
infor­ma­
tion."},
{"role": "user", "con­ tent": "John and Susan are
going to an AI con­ fer­
ence on Fri­ day."},
],
response_for­mat=Cal­en­
darEvent,
)

print(com­
ple­
tion.choices[0].mes­
sage.parsed)

35
JavaScript

import Ope­
nAI from "ope­
nai";
import { zodRespon­
se­
For­
mat } from "ope­
nai/helpers/zod";
import { z } from "zod";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai"
});

const Cal­en­
darEvent = z.object({
name: z.string(),
date: z.string(),
par­
tic­
i­
pants: z.array(z.string()),
});

const com­ple­
tion = await ope­nai.beta.chat.com­
ple­
‐
tions.parse({
model: "gemini-​2.0-​
flash",
mes­
sages: [
{ role: "sys­tem", con­
tent: "Extract the event infor­
ma­‐
tion." },
{ role: "user", con­tent: "John and Susan are going to
an AI con­
fer­ence on Fri­
day" },
],
response_for­mat: zodRespon­
se­For­
mat(Cal­
en­
darEvent,
"event"),
});

const event = com­

ple­
tion.choices[0].mes­
sage.parsed;
con­
sole.log(event);

36
Embed­dings
Text embed­dings mea­sure the relat­ed­ness of text strings and can be
gen­er­ated using the Gem­ini API → https://ai.google.dev/gemini-api/docs/em
beddings.

Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

response = client.embed­
dings.cre­
ate(
input="Your text string goes here",
model="text-​
embedding-004"
)

print(response.data[0].embed­
ding)

37
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
});

async func­
tion main() {
const embed­
ding = await ope­
nai.embed­
dings.cre­
ate({
model: "text-​
embedding-004",
input: "Your text string goes here",
});

con­
sole.log(embed­
ding);
}

main();

REST

curl "https://gen­
er­
a­tive­
lan­‐
guage.googleapis.com/v1beta/ope­ nai/embed­
dings" \
-H "Content-​
Type: appli­ca­tion/json" \
-H "Autho­
riza­
tion: Bearer GEM­ INI_API_KEY" \
-d '{
"input": "Your text string goes here",
"model": "text-​
embedding-004"
}'

38
extra_body
There are sev­eral fea­tures sup­ported by Gem­ini that are not avail­‐
able in Ope­nAI mod­els but can be enabled using the extra_body
field.
extra_body fea­tures

safety_set­
‐ Cor­re­sponds to Gem­ini's Safe­ty­Set­ting.
tings

cached_con­
tent Cor­re­sponds to Gem­ini's Gen­er­ate­Con­tentRe­‐
quest.cached_con­ tent.

think­
ing_con­
‐ Cor­re­sponds to Gem­ini's Think­ing­Con­fig.
fig

cached_con­
tent
Here's an exam­ple of using extra_body to set cached_con­tent:

39
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key=MY_API_KEY,
base_url="https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/"
)

stream = client.chat.com­ple­tions.cre­
ate(
model="gemini-​2.5-pro",
n=1,
mes­
sages=[
{
"role": "user",
"con­
tent": "Sum­ma­
rize the video"
}
],
stream=True,
stream_options={'include_usage': True},
extra_body={
'extra_body':
{
'google': {
'cached_con­
tent': "cached­Con­
‐
tents/0000aaaa1111bbbb2222cccc3333dddd4444eeee"
}
}
}
)

for chunk in stream:

print(chunk)

40
 print(chunk.usage.to_dict())

List mod­els
Get a list of avail­able Gem­ini mod­els:

Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

mod­
els = client.mod­
els.list()
for model in mod­
els:
print(model.id)

41
JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/",
});

async func­
tion main() {
const list = await ope­
nai.mod­
els.list();

for await (const model of list) {

con­
sole.log(model);
}
}
main();

REST

curl https://gen­
er­
a­
tive­
lan­
guage.googleapis.com/v1beta/ope­
‐
nai/mod­
els \
-H "Autho­
riza­
tion: Bearer GEM­
INI_API_KEY"

Retrieve a model
Retrieve a Gem­ini model:

42
Python

from ope­
nai import Ope­
nAI

client = Ope­
nAI(
api_key="GEM­
INI_API_KEY",
base_url="https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/"
)

model = client.mod­
els.retrieve("gemini-​
2.0-​
flash")
print(model.id)

JavaScript

import Ope­
nAI from "ope­
nai";

const ope­
nai = new Ope­
nAI({
apiKey: "GEM­
INI_API_KEY",
baseURL: "https://gen­
er­
a­
tive­
lan­
‐
guage.googleapis.com/v1beta/ope­
nai/",
});

async func­
tion main() {
const model = await ope­
nai.mod­
els.retrieve("gemini-​
2.0-​
flash");
con­
sole.log(model.id);
}

main();

43
REST

curl https://gen­
er­
a­tive­
lan­
guage.googleapis.com/v1beta/ope­
‐
nai/mod­
els/gemini-​
2.0-​flash \
-H "Autho­
riza­
tion: Bearer GEM­ INI_API_KEY"

Cur­rent lim­it­ a­tions

Sup­port for the Ope­nAI libraries is still in beta while we extend fea­‐
ture sup­port.
If you have ques­tions about sup­ported para­me­ters, upcom­ing
fea­tures, or run into any issues get­ting started with Gem­ini, join
our Devel­oper Forum → https://discuss.ai.google.dev/c/gemini-api/4.

What's next
Try our Ope­nAI Com­pat­ib
­ il­ity Colab → https://colab.sandbox.google.com/
github/google-gemini/cookbook/blob/main/quickstarts/Get_started_OpenAI_Compat
ibility.ipynbto work through more detailed exam­ples.
Except as oth­er­wise noted, the con­tent of this page is licensed under
the Cre­ative Com­mons Attri­bu­tion 4.0 License → https://creativecommon
s.org/licenses/by/4.0/, and code sam­ples are licensed under the Apache
2.0 License → https://www.apache.org/licenses/LICENSE-2.0. For details, see
the Google Devel­op­ers Site Poli­cies → https://developers.google.com/site-p
olicies. Java is a reg­is­tered trade­mark of Ora­cle and/or its affil­ia
­ tes.
Last updated 2025-06-18 UTC.

44