Full Transcript

THIS FREE AI Tool Is Better Claude Code! (Google Gemma 4)

8:24906 summary words · ~5 min readEnglishBy Ishan SharmaTranscribed Jul 19, 2026

Summary

Google's free, local Gemma 4 model paired with LM Studio and the Continue VS Code extension delivers Claude Code-level coding assistance at zero cost and with no usage limits.

Eliminating per-use fees and reset waits removes the main barrier that stops solo developers from shipping AI-generated apps continuously.

Section summaries

Claude Code Pain and Gemma 4 Pitch

watch

The host opens by calling Claude Code the top AI coding agent of 2026 but flags its cost and usage caps that force 5-hour waits. He introduces Gemma 4 as a free, offline, unlimited alternative that matches Claude at writing code and promises a walkthrough of using it as a file-creating assistant.

Claude Code is excellent but expensive with hard usage resets
Gemma 4 is free, local, and claimed equal in code quality

Sets the problem and thesis for the whole video.

Feature Comparison and Whisper Flow Ad

optional

He shows a generated infographic comparing Gemma 4 with Claude Opus 4.8 across raw code quality, reasoning depth, agentic workflows, and privacy, noting parity except for $0 cost. He then pivots to a sponsor: Whisper Flow, a phone app that listens once to a spoken prompt and outputs a cleaned prompt to avoid context breakage in typed sessions, demonstrated with a solo-creator tool research request.

Comparison shows Gemma 4 free with similar capability scores
Whisper Flow converts voice to structured prompts to cut reprompting

Comparison is superficial and the middle is a paid sponsor demo.

Whisper Flow Demo

skip

The host opens his phone, enables Whisper Flow, and speaks a detailed prompt asking for research on 2026 AI tools for solo content creators with a listicle and business idea. The app returns a cleaner prompt than his speech, which he credits for preventing wasted credits and reprompting, then points to the description link.

One-shot voice capture yields better prompt text than raw speech
Link to try Whisper Flow is in the description

Pure sponsor segment unrelated to the Gemma 4 setup.

Installing Gemma 4 in LM Studio

watch

He instructs viewers to download LM Studio as the GUI for local models, search and install Gemma 4 (the 26B param version he already loaded), select it, and load it for chat. He tests with a Python snake game request and notes the model writes code but cannot run it or manage files yet, acting like a basic chat.

LM Studio is the interface to run Gemma 4 locally
Base LM Studio chat cannot create files or execute code

Core first step of the local agent setup.

Continue Extension in VS Code

watch

To overcome chat limits, he opens VS Code, goes to extensions, and installs Continue, which links local LLMs as coding assistants. He opens a new 'Gemma 4' folder and explains that before using Continue you must return to LM Studio and click start server on port 123 to expose the model.

Continue extension bridges local models to VS Code
LM Studio server must run on port 123 for the connection

Defines the critical bridge from chat to agent.

Connecting and Configuring Continue

watch

With the server up, he adds a chat model in Continue, picks LM Studio, autodetects, and sees Gemma 4 26B loaded. He opens Continue settings to show built-in tools: read file automatically, create new file ask-first, run command ask-first, and advises granting the permissions you want before coding.

Autodetect in Continue surfaces the local Gemma 4 model
Tool permissions control file and command autonomy

Shows the exact config needed for agentic behavior.

Building a CRUD App and Claude Compare

watch

He prompts Continue to build a to-do CRUD app in HTML with add/delete/view/update saved to a file and run; Gemma 4 creates the directory and file, and Go Live shows a working list. He runs the same prompt in Claude Code, shows a similar result, and claims Gemma 4's UI looks more presentable, leaving final judgment to viewers.

Gemma 4 via Continue produced a runnable CRUD app from one prompt
Claude Code output was functionally similar in the demo

The pivotal proof of equal practical capability.

Landing Page Build and Wrap-Up

optional

He tells the agent to discard the CRUD code and build a single-file protein brand landing page with header, copy, testimonials, and differentiators; the agent enters agent mode, thinks, and writes the new file. He notes plan/chat mode switches, recaps the VS Code agent setup, points to the full guide and Whisper Flow link in the description, and signs off.

Agent mode handled a full page rewrite in one file
Modes include agent, plan, and chat inside Continue

Additional example and outro with no new setup info.

Key points

Free Local Alternative to Claude Code — Gemma 4 is an open Google model (26B params) that matches Claude Opus 4.8 on code quality and agentic workflows while costing $0 and running offline.
Local Stack Setup — You download LM Studio, load Gemma 4, start its server on port 123, then connect it inside VS Code via the Continue extension to get file creation and command execution.
Voice Prompting Reduces Context Loss — The sponsor Whisper Flow captures a one-shot spoken prompt and cleans it into a structured text prompt, avoiding the broken context common with typed Claude sessions.

“But the problem is it's very expensive to run and B you can run out of your clawed usage and then you are sitting for 5 hours waiting for the usage to be reset.” — Host

“It is a 26 billion parameter model that I installed a few days ago.” — Host

AI-generated from the transcript. May contain errors.

0:00

Claude Code is the number one AI coding

0:02

agent in 2026. We have all tried it out.

0:05

You know that it's amazing at building a

0:08

plan, building all the code, creating

0:10

all the files and creating that final

0:12

app for you step by step. It's

0:14

exceptional at all of that. But the

0:16

problem is it's very expensive to run

0:18

and B you can run out of your clawed

0:20

usage and then you are sitting for 5

0:22

hours waiting for the usage to be reset.

0:24

But now imagine a coding assistant that

0:26

works for completely free. You can use

0:28

it as much as you want without access to

0:32

internet. It's called Gemma 4 and it is

0:34

just as good as clot code when writing

0:36

code. And today I'm going to show you

0:38

how it can become your AI coding

0:40

assistant writing all the files for you,

0:42

creating all the apps that you want. So

0:44

this right here is cloud code. You can

0:45

access it through the terminal. You can

0:47

use it independently in the cloud app or

0:49

you can just use it as an extension

0:51

inside of Visual Studio Code. On the

0:53

other hand, we have Gemma 4, an open

0:56

model by Google that you can download

0:58

and just use it wherever you are,

1:00

whenever you want, for as much as you

1:02

want. Let me show you the differences

1:04

and the comparison between the both. Can

1:07

you make an infographic comparing Gemma

1:09

4 with Claude Opus 4.8 and Claude code

1:13

in general and show its capabilities and

1:16

which is better. And let's look at the

1:18

comparison now. So we have raw code

1:21

quality, reasoning depth, agentic

1:23

workflows, privacy, and you can see that

1:25

it's actually very good, just as good as

1:27

Claude Opus 4.8 in a lot of the tasks.

1:30

But the best thing is ongoing cost

1:32

efficiency because this is completely

1:33

free. So you pay $0 and you can use it

1:36

as much as you want. Now, there's one

1:38

problem that a lot of you will get into.

1:40

Most people are typing prompts into

1:42

Claude. It's not wrong, but you can lose

1:44

your train of thought easily. the

1:45

context can break and you end up getting

1:47

a bad lowquality output and that's

1:49

exactly why I've been using whisper

1:51

flow. Let me show you with a simple

1:52

example. So I will open up my phone and

1:55

enable whisper flow. Okay, so I want you

1:58

to research the top AI tools that are

2:00

useful for content creators for solo

2:02

content creators specifically. Research

2:04

for the tools that are actually being

2:06

used in 2026 to make money or to build

2:09

something. It should cover different use

2:11

cases like brainstorming or ideating

2:13

content, making automations, video

2:15

editing, designing and so on and then

2:17

present it as a listical to me along

2:20

with a solid business idea that can be

2:22

experimented with those tools and just

2:25

look at this. It listen to the prompt

2:27

only once and just see how much better

2:30

the prompt is than what I just spoke and

2:32

that's the amazing part about this tool.

2:33

It understand exactly what you mean the

2:35

first time so that you're not

2:36

reprompting or wasting credits. Check

2:38

the link in the description to try out

2:40

Whisper Flow and let's get back to the

2:41

video. Okay, let's get started with

2:43

Gemma 4. First of all, you need to

2:44

download LM Studio. So, this is the

2:46

graphical user interface that will allow

2:48

you to interact with this new local

2:50

model. Okay, this right here is LM

2:52

Studio. Once you download, you have to

2:54

simply now install the Gemma 4 model.

2:56

For that, you can simply go over here

2:58

and search for Gemma 4 and then just

3:00

install it. I already have this

3:01

installed. It is a 26 billion parameter

3:03

model that I installed a few days ago.

3:06

And once you download, you simply have

3:07

to go to pick a model and select Gemma 4

3:11

26 billion parameter model. And we'll

3:13

just say load this model. So now I can

3:17

simply just chat in LM Studio and use

3:20

Gemma 4 as my large language model to

3:23

answer all of my queries. Once this

3:25

loads up and the model is loaded, so I

3:27

can ask a very simple question uh write

3:29

a simple snake game in Python

3:34

and it will now start writing all the

3:36

code for me. But there's one problem

3:39

with it. You can't really run the code

3:41

right now. Um or it only generates the

3:43

code but doesn't create files, doesn't

3:46

think through the entire project and it

3:48

just works as a simple chat just like

3:50

chat GPT. There's one way to change that

3:53

and for that we will have to open up

3:55

Visual Studio Code. So let's open up

3:58

Visual Studio

4:01

and here is where you need to go to

4:03

extensions and click on continue. So

4:07

continue is a extension that allows you

4:09

to access your local large language

4:11

model and use that as your AI coding

4:14

assistant. So I'll simply go on to

4:16

continue and I will install it. I've

4:19

already installed it right now. So I can

4:21

just access it by going on to continue

4:24

right here. So I can be in my folder.

4:27

Let me open a new folder over here.

4:30

Let's go here. Call new folder. Let's

4:32

just call this Gemma 4. Open this. And

4:36

before you use continue, you have to go

4:37

over here to LM Studio and click on

4:39

start server on port 123. So basically

4:42

you are now running your Gemma 4 model

4:46

on this server port. And once you do

4:49

that, you can now simply go on to

4:51

continue and click on add chat model.

4:54

Here you will find different providers.

4:56

So you have anthropic, you have openAI,

4:58

you have so many of them that you can

5:00

click add your API key and start using.

5:03

But we wanted to use LM studio and we

5:06

wanted to autodetect our models and

5:08

click on connect. And when you do that,

5:10

we can now see Gemma 4 26 billion

5:13

parameter model loaded right here inside

5:16

of Visual Studio Code. All with the help

5:19

of continue. Before we use it to write

5:21

any code, you have to go into the

5:22

settings and actually check what it can

5:26

really do. So built-in tools read file

5:28

automatically. Create new file ask

5:30

first. Run Tamil command ask first. So

5:32

you can give it all the permissions that

5:34

you want it to have. But once we've

5:35

enabled that, we can simply just ask it

5:37

to create a simple file. So I'll just

5:39

say

5:41

create a simple CRUD app in which I can

5:45

add tasks to my to-do list. I can

5:47

delete, I can view, and I can also

5:49

update certain tasks that I want to in a

5:52

simple HTML page and save all of this to

5:55

a file and help me run the file as well.

5:59

And now I can just click on enter. And

6:01

what you will see is it will start

6:03

writing all the files. It will start

6:05

creating a simple directory and then

6:07

we'll be able to run the CRUD app as

6:09

well.

6:14

And there you go. It wrote an entire

6:16

script and it also created a new file

6:19

into our folder that I can simply run.

6:22

So if I go onto here, I see a new file

6:24

being made here. And now I can just

6:25

click on go live. And this is our to-do

6:29

list. So I can say uh gimming. I can add

6:32

it as a task. I can say uh swimming add

6:35

that as a task. I can also delete this

6:38

if I want to. I can also edit this uh

6:41

protein

6:43

and as you can see it just works really

6:45

really well. So that's what you can just

6:47

do very simply with the help of Gemma 4.

6:49

Now the same prompt on cloud code and

6:52

this is what it created. I think it

6:54

works very simply just like the one that

6:56

we've seen but this is what it looks

6:58

like. I let you decide if this looks

7:00

good or the one that Gemma 4 made looks

7:03

better. So, we can do that. We can

7:05

delete this if you want. And it just

7:06

looks very similar. I still think that

7:09

the Gemma 4 one looks more presentable.

7:11

But we can just keep building from here

7:12

onwards. I can just say, "Okay, get rid

7:15

of all of this code. And now build me a

7:17

landing page for a protein brand and

7:21

make sure that you have a nice header,

7:24

that you have a great copy. Ensure that

7:26

you show testimonials by people and also

7:29

make sure that you showcase why this

7:32

protein is better than the other

7:33

options. Do it all in just one single

7:37

file.

7:39

Let's say that's what I want to do. And

7:41

then I'll press on enter. You can

7:43

already see it thinking through this

7:45

problem. So it's getting rid of all of

7:46

the logic it wrote for the to-do list

7:48

app. And it is now writing new code for

7:50

this landing page for this protein

7:52

brand. You can also switch between

7:53

different modes. So we currently have

7:55

the agent mode on. But we can also use

7:57

plan or chat mode over here. But this

7:59

was my comparison of Gemma 4 and claude

8:02

code. We saw how to use Gemma 4 on

8:04

Visual Studio Code as a coding agent to

8:07

create anything that you wanted to.

8:09

Complete guide to using Gemma 4 is in

8:11

the description of this video along with

8:13

the whisperflow link. So check it out,

8:16

install it and start prompting AI

8:18

better. I hope this helps. I'll see you

8:20

in the next one. Bye-bye.

More transcripts

Explore other videos transcribed with YouTLDR.

Get the TLDR of any YouTube video

Process a new video with YouTLDR Pro — transcripts, summaries, translations, and repurposing in 125+ languages.

Get YouTLDR Pro — $12/month