Claude Pro vs GPT Plus: $20 Plans Reviewed for Professional Use

Name: Claude Pro vs Codex: Is $20 Plan Enough?
Uploaded: 2026-03-30T17:09:39.832085+00:00
Duration: 8 min 25 s
Channel: Eric Michaud
Description: Summary and key takeaways on Claude Pro vs Codex: Is $20 Plan Enough?: Summary & Key Takeaways, covering Subscription Comparison Both Claude Pro and GPT Plus

Eric Michaud

Mar 30, 2026

•

8 min video

•

2 min read

YouTube video ID: JjP6NIRJeDQ

Source: YouTube video by Eric Michaud — Watch original video

PDF

Both Claude Pro and GPT Plus cost $20 per month, making them directly comparable entry‑level offers. Claude Pro provides access to Claude Code and the Opus 3.5/Sonnet 3.5 models, while GPT Plus unlocks the Codex 5.4 flagship model. The comparison focuses on the size of each plan’s “gas tank” rather than raw horsepower.

Usage and Rate Limits

Claude Pro allocates 44,000 tokens for every 5‑hour window. The window refreshes from the moment the first prompt is sent, preventing a user from exhausting the entire weekly allowance in a single session. Sonnet serves as the baseline model; Opus consumes usage about 70 % faster, and Haiku can be used to stretch capacity for lighter tasks. The speaker’s data estimate the API‑equivalent cost of the $20 plan at roughly $180 per month.

GPT Plus limits the flagship Codex model to roughly 160 messages every three hours. If the limit is reached, the system nudges the user toward a smaller model or the API, maintaining workflow continuity. Codex API costs are significantly lower than Claude’s API costs, and the speaker’s usage data show daily token consumption of 50,000–100,000, with occasional spikes to 700,000.

Strategic Implementation

Claude excels at planning, workflow design, and activity logging, so it is used for everyday tasks on the left side of the workflow. Codex shines at executing directions, making it the right‑hand tool for task completion. Conscious model switching sustains the $20 plans: Sonnet is the default, Opus handles heavy builds, and Haiku stretches the “gas tank” for planning phases. Balancing Claude and Codex in tandem avoids hitting usage caps while maximizing efficiency.

Conclusion

The $20 subscription tier remains sustainable for professional work when models are selected strategically. Claude Pro’s token usage translates to $8–$14 per day in API‑equivalent costs, and Codex’s lower API expenses keep overall spending within the budget. Proper model selection and awareness of rate limits enable continuous productivity without exceeding the entry‑level plans.

Takeaways

Claude Pro and GPT Plus both cost $20 per month and are evaluated based on their token or message “gas tank” capacity rather than raw model horsepower.
Claude Pro grants 44,000 tokens per 5‑hour window, with Sonnet as the baseline model, Opus consuming usage 70 % faster, and Haiku extending capacity for lighter tasks.
GPT Plus limits the flagship Codex model to roughly 160 messages every three hours and automatically rolls users to a smaller model or the API when the limit is reached.
Effective workflow strategy pairs Claude for planning and logging while using Codex for task execution, switching models consciously—Sonnet for default, Opus for heavy builds, Haiku for planning—to stay within the $20 plan limits.
Based on professional usage data, the $20 plans remain sustainable, with Claude Pro’s API equivalent costing $8‑$14 per day and Codex’s API costs staying significantly lower.

Frequently Asked Questions

How does Claude Pro calculate its token allowance?

Claude Pro allocates 44,000 tokens for each 5‑hour window, resetting the window from the moment the first prompt is sent. This design prevents users from depleting the entire weekly allowance in a single session while keeping usage within the $20 plan.

What happens when GPT Plus reaches its message limit?

When the 160‑message limit for the Codex flagship model is hit, the system nudges the user to a smaller, less intensive model or directs them to the API. This automatic roll‑down maintains access while respecting the constraints of the $20 subscription.

Who is Eric Michaud on YouTube?

Eric Michaud is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.

Does this page include the full transcript of the video?

Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.

Helpful resources related to this video

If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.

Ergonomic Dual Monitor Desk Mount Stand Recommended

Provides the physical screen real estate needed to keep Claude and Codex open side-by-side for efficient workflow management.

Amazon →

Mechanical Keyboard With Programmable Macro Keys

Allows for faster input and custom shortcuts to switch between AI interfaces or trigger common prompt templates.

Amazon →

Large Desk Mat For Keyboard And Mouse

Provides a stable, comfortable surface for long hours of coding and AI-assisted development work.

Amazon →

Links may be affiliate links. We only include resources that are genuinely relevant to the topic.

Summarize another video

Full Transcript YouTube

In this video, we're going to talk about
whether the $20 plan of Codex and Cloud
Pro are enough to actually work 7 days a
week. And as much as I'd love for you to
watch the whole video, short answer is
if you have to ask, probably is, cuz
you're probably not doing like super
token nuke style builds anyway. Except
that answer is super loaded. It really
depends on what model you're using, what
you're actually building, and if you
have any guard rails in place to keep
the model from thinking its way into
oblivion. The thing is, I've actually
got the numbers. This video has been a
bit coming. I've been running on the $20
plan of Cloud Pro and Codeex for the
last two weeks. I can actually answer
the question. So, we're going to look at
where I was rate limited, what kind of
work got me there and what you can
realistically expect to get done with
either model before each tells you to
take a break. I'm comparing apples to
apples here. These are the very baseline
entry level paid offers for each
platform. Okay, I'm talking $20 Claude
Pro plan and $20 GPT plus. It's the
entry level for where you get cla uh
sorry cloud code and access to opus 4.6
and on GPT plus you get codeex app and
you're able to access uh codeex 5.4 in
the terminal which is the one I was
using. I am not talking about which
one's smarter, which one's better. I'm
not going down that rabbit hole today.
They're comparable enough and it's more
about comparing gas tank rather than
horsepower. Like does this work for you
as a professional and what you can
expect. Okay, my use case is typical
agency business owner kind of stuff. I
was using it for building scripts,
editing files, agent workflows,
automations. I had one large client
build that I did end to end this week as
well. So, let's get into it. So, check
this out. Today's the day we're
recording because today is the day my
credits refreshed. We've got 7 days
worth of usage. I used extra usage to
compile the notes from the last week.
All of my different tasks and token
counts together. This is actual use case
data. All right. So, just to start with
cloud since that's what I've got open.
The basic breakdown of how the usage
limit is calculated is you've got a
certain amount of compute for the week
and you've got a certain amount of
tokens for every 5 hour session just to
throttle you so you're not going through
your entire week's usage in one session.
How that works out is you get 44,000
tokens per 5 hour window refreshing as
you go from your first prompt. So, if
you went to the API and ran Sonnet for
yourself, you're paying about $3 per
million input tokens and $15 per million
output tokens. Which brings me to an
important point. you should always count
sonnet as your baseline. If you're on
the $20 plan, just don't use opus unless
it's something heavy like an actual
build or or whatever like that. It's
it's not a good idea. I'm going to get
more into that later, but you you'll see
what I mean. Okay, so we're going with
the assumption that sonnet 4.6 is your
baseline here. Anthropic's own
guidelines says the average cloud code
developer uses about $6 to 12 in AP API
equivalent credits a day. Spoiler alert,
I used more than that. So they suggest
on that it looks like about $180 worth
of API equivalent usage if you're going
through the $20 plan on Sonnet. So from
their numbers still looks like a great
deal, right? According to Cloud's own
documentation and all of the different
sources I was able to find here, the pro
tier is for people using Claude every
day with for moderate intensity tasks,
right? And that is based on Sonnet.
Okay, so Opus is going to burn your
usage 70% faster. On the other side of
that coin, if you switch to Haiku for
like lighter tasks, you're also going to
stretch your usage by quite a bit as
well. So anyway, my actual usage, here's
what I actually did on Claude Pro this
week based on all the task completions
that I did in this vault. And I used
between 2.5 and 3.5 million tokens with
my Claude subscription. Specifically,
the Sonnet equivalent pricing on API,
that's about $8 to $14 a day. My cloud
subscription is paying for itself daily.
And I had one large client project that
I pushed through at about 1 million
tokens in like one shot and I hit the
session limit, but I was able to finish
it up a little later. And uh that so
like I said, this is actually like a
really good litmus test for real life
usage. One large client project a week.
So I mean when you think about it,
Claude gives you a ton for $20, right?
And at the sticker price, I'm getting
significantly more value for the dollar
if I go with a subscription than I would
be if I was like paying for every API
call. Now, flipping over to chat GBT
plus, which is what gets you codeex 5.4.
So, right now, based on everything I
could find, it says that the plus tier
gives you about 160 messages every 3
hours with the top GPT 5 whatever model.
I was using 5.4 and then it would nudge
you down to like a a smaller model for
everyday use or you could turn over into
the API. Right? So, it works a little
bit differently than cloud code because
codeex 5.4 is their flagship model,
right? like it is kind of like using
Opus for everything and there is
something to think about for sure. The
only thing is I've never actually hit
that rate limit on Codeex. So I can't
speak exactly to what happens when you
do. Um it does say that you roll down to
a smaller model or you can get into the
API. And that's another thing worth
noting is that the Codeex API cost is
really inexpensive, like substantially
less expensive than Claude at the
moment. So, for my actual Codex numbers,
I didn't use anywhere near as many
tokens as I did with Claude. But that's
also because Codex doesn't use as many
tokens as the anthropic models do, but I
used 50 to 100,000 tokens a day. Um,
except for one specific day where I
actually bumped it up to 700,000 um for
one larger build, but I got it done in
the one shot and didn't trip rate limits
there either. The nice thing about this
is that that is still really inexpensive
if even if I was to go on the Codeex API
credits. So, from my logs, my actual
journal and list of tasks that I
actually completed, Claude Code gave me
about 3.5 million tokens last week, and
I brushed up about 85% of my weekly bar.
This guy here, this one, but I never
fully hit it. And Codex 5.4 let me run
700,000 token build and sit in it, you
know, throughout the week. And I didn't
ever have to stop. And the API math is
like even if I did and I went into extra
usage, which is what I can do on cloud
as well, I it would have cost me almost
no money at all. Day-to-day, my layout
is pretty much exactly this where I have
cloud on the left doing my everyday sort
of tasks, running my workflows, logging
my activities, you know, I've got it
running hooks and different automations
in the background, and then I'll have
Codex on the right that just kind of
goes through and crosses stuff off the
list. Codeex is fantastic at following
directions. Claude is really good at
planning stuff and coming up with things
for codecs to do. I'll often very much
run these in tandem, but I'm confident
the reason I was able to get this much
use out of Claude was because I was
conscious of which model I was using for
what task. I'm always on sonnet medium
effort for that's like the the default
the fallback. The the factory settings
are great for extended use. And I do go
to Opus if I'm actually doing the
builds. Also bouncing down to high coup
for more back and forth planning things.
There are definitely instances where I
realize I need to like give it a little
bit more horsepower to complete the task
and that's okay. Whereas Codex is all
gas, no brakes for my wrestling fans. If
you're uh if you're using that, you're
always using Codex 5.4. It's just a
really efficient tool. I like it a lot.
So, to answer the overarching question
of the entire video, can the $20 plan
sustain a week of work? Yes, absolutely.
On either model, you can absolutely do
that. On Claude though, you need to be a
little bit conscious of what it is that
you're doing. If you had 20 bucks and
you had to choose just one, if you're
doing like office admin and basic builds
like like Claude says, moderate
intensity tasks and you stick to Claude
Sonnet, Cloud Pro has a lot going for it
out of the box like you cloud code like
co-work channels, other different
plugins and support, the desktop app,
everything, it's definitely worth it.
But I mean, if you absolutely don't
trust yourself from uh bumping it into
Opus for like everything, maximum
horsepower for the dollar, Codeex is the
safer play for sure if you had to get
just one. its practical rate limits are
a lot looser. And if you do wind up
hitting a wall, the API is a lot less
expensive than Claude. You can continue
on extra usage and not have to worry
about breaking the bank. Am I going to
bounce up to the $100 Cloud plan? You
know what? Surprisingly, maybe not.
Like, probably will. But, um, doing
these two things side by side, I I
didn't run into any problems not
changing my actual output on anything.
You know, I was still working the same
amount. So, I think the the best is the
$40 plan that you build yourself. Okay.
So hopefully that helped. I've had the
fortunate opportunity to talk to like a
hundred people in the YouTube comments
about this exact question over the last
couple weeks. Make sure to check out
this video to see how I set up my
workspace and run my business dayto-day.
Thank you for hanging out. We'll catch
you in the next one.