·HSK 4

Google (technology company)
Gǔgē
谷歌
to release / publish
fābù
发布
three (models/types)
sān kuǎn
三款
new
xīn
新
Gemini (Google AI model family)
Gemini
Gemini
AI model(s)
móxíng
模型
，
flagship version
qíjiàn bǎn
旗舰版
coding / programming
biānmǎ
编码
capability / performance
nénglì
能力
significantly / by a large margin
dàfú
大幅
improved / enhanced
tíshēng
提升
and / also
bìng
并
to reduce / lower
jiànglì
降低
cost
chéngběn
成本
。

Google

Gǔgē

谷歌

on / at (time preposition)

yú

于

July 21

7 yuè 21 rì

7月21日

released / published

fābù

发布

three (models)

sān kuǎn

三款

Gemini

new AI models

xīn móxíng

新模型

，

flagship version

qíjiàn bǎn

旗舰版

Gemini 3.6 Flash (Google AI model)

Gemini 3.6 Flash

programming / coding

biānchéng

编程

capability

nénglì

能力

significantly / by a large margin

dàfú

大幅

improved / enhanced

tíshēng

提升

，

output

shūchū

输出

cost

chéngběn

成本

simultaneously / at the same time

tóngbù

同步

declined / dropped

xiàjiàng

下降

。

new version(s)

xīn bǎnběn

新版本

also / additionally

hái

还

includes / features

bāokuò

包括

specifically for / designed for

zhuānwèi

专为

cybersecurity

wǎngluò ānquán

网络安全

vulnerabilities / security flaws

lòudòng

漏洞

detection / identification

jiǎncè

检测

designed / purpose-built

shèjì de

设计的

Gemini 3.5 Flash Cyber (cybersecurity AI model)

Gemini 3.5 Flash Cyber

，

currently / at present

mùqián

目前

limited to / only available to

jǐn xiàn

仅限

certain / some

bùfen

部分

government agencies / institutions

zhèngfǔ jīgòu

政府机构

to use / for use

shǐyòng

使用

。

July 25, 2026 at 8:45 AM

Google (technology company)

Gǔgē

谷歌

on / at (time preposition)

yú

于

July 21

7 yuè 21 rì

7月21日

released / published

fābù

发布

(completion particle)

了

three (types/models)

sān kuǎn

三款

new

xīn de

新的

Gemini (Google AI model family)

Gemini

artificial intelligence

réngōng zhìnéng

人工智能

models

móxíng

模型

。

flagship version

qíjiàn bǎn

旗舰版

Gemini 3.6 Flash (Google AI model)

Gemini 3.6 Flash

zài

在

programming capability / coding performance

biānchéng nénglì

编程能力

in terms of / in the area of

fāngmiàn

方面

achieved / made

qǔdé le

取得了

significant progress / major gains

zhòngdà jìnzhǎn

重大进展

，

zài

在

DeepSWE (coding benchmark test)

DeepSWE

benchmark test / evaluation

jīzhǔn cèshì

基准测试

in / within

zhōng

中

score

défen

得分

from

cóng

从

37 percent

37%

improved to / rose to

tíshēng zhì

提升至

49 percent

49%

。

this model

gāi móxíng

该模型

also / additionally

hái

还

(disposal preposition)

jiāng

将

output tokens

shūchū lìngpái

输出令牌

(particle)

的

usage volume / consumption

shǐyòng liàng

使用量

reduced / decreased

jiǎnshǎo

减少

(completion particle)

了

approximately / about

yuē

约

17 percent

17%

，

output pricing

shūchū dìngjià

输出定价

from

cóng

从

per million tokens

měi bǎiwàn lìngpái

每百万令牌

9 měiyuán

9美元

reduced to / lowered to

xiàtiáo zhì

下调至

$7.50

7.5 měiyuán

7.5美元

。

Gemini 3.6 Flash

supports / is compatible with

zhīchí

支持

up to / maximum

zuìgāo

最高

one million tokens

100 wàn lìngpái

100万令牌

(particle)

的

context window

shàngxiàwén chuāngkǒu

上下文窗口

，

zài

在

computer use / computer operation

jìsuànjī cāozuò

计算机操作

test / benchmark

cèshì

测试

in / within

zhōng

中

score

défen

得分

also

yě

也

showed / had

yǒu

有

notable / significant

míngxiǎn

明显

improvement / gain

tíshēng

提升

。

this time / in this release

cǐ cì

此次

also / additionally

hái

还

released

fābù

发布

(completion particle)

了

aimed at / for

miànxiàng

面向

high throughput

gāo tūntǔ liàng

高吞吐量

low latency

dī yánchí

低延迟

tasks

rènwu

任务

(particle)

的

Gemini 3.5 Flash-Lite (lightweight AI model)

Gemini 3.5 Flash-Lite

。

the third / third type

dì sān kuǎn

第三款

model

móxíng

模型

Gemini 3.5 Flash Cyber (cybersecurity AI model)

Gemini 3.5 Flash Cyber

specifically for / designed for

zhuānwèi

专为

to detect / identify

jiǎncè

检测

software

ruǎnjiàn

软件

security vulnerabilities

ānquán lòudòng

安全漏洞

designed for this purpose

ér shèjì

而设计

，

currently / at present

mùqián

目前

limited to / only available to

jǐn xiàn

仅限

governments

zhèngfǔ

政府

and

hé

和

specific / designated

tèdìng

特定

partner organizations / collaborating institutions

hézuò jīgòu

合作机构

to use / for use

shǐyòng

使用

。

three models

sān kuǎn móxíng

三款模型

(particle)

的

release / launch

fābù

发布

further / additionally

jìn yī bù

进一步

intensified / heightened

jiājù

加剧

(completion particle)

了

Google

Gǔgē

谷歌

and / with

yǔ

与

other

qítā

其他

major / leading

zhǔyào

主要

AI labs / AI research organizations

AI shíyànshì

AI实验室

zài

在

model performance

móxíng xìngnéng

模型性能

and

hé

和

cost

chéngběn

成本

in terms of / in the area of

fāngmiàn

方面

(particle)

的

competition / competitive race

jìngzhēng

竞争

。

News in English

Science & Technology

Google Releases Three New Gemini Models With Flagship Gemini 3.6 Flash Showing Major Coding Gains and Lower Costs

Google released three new Gemini AI models on July 21, led by Gemini 3.6 Flash, which improved coding scores by 12 percentage points and cut output costs by 17 percent. A specialized cybersecurity variant, Gemini 3.5 Flash Cyber, is available only to select government partners.

News in English

Google Releases Three New Gemini Models With Flagship Gemini 3.6 Flash Showing Major Coding Gains and Lower Costs

More in Science & Technology