- Joined
- Mar 29, 2025
What are you trying to use Claude for? For light programming, a Pro plan is okay, but you'll likely hit 5h limits pretty quickly.
Claude AI discussion
- Thread starter mister fireworks
- Start date
Kether
kiwifarms.net
- Joined
- Aug 8, 2025
Sam Altman knows how to talk to investors, Dario doesn't. when you get to c-suite its less about competency and more about schmoozing.
nope, pro also has limits and constraints. they just take 5x longer to hit.
i'm actually pretty shocked null hasn't complained about the limits. that's usually people's biggest complaint, especially with how much it feels like a bait and switch to many users
- Joined
- Oct 19, 2024
When you turn down Mansa Musa-tier generational wealth just to preserve Super Happy Shrimp Maximalization Funtime with your polycule in San Fransisco
- Joined
- Mar 17, 2022
I've been using Claude as a search engine more than a writer of finished code and I haven't run out of the free plan yet; granted, I'm just messing with pyqt6 so it's nothing esoteric or low-level that I'm prompting it about, but I'm still impressed at how accurate all of its recommendations have been. In short, a helpful digestion of documentation and StackOverflow that vastly reduces my search time for explanations and clarifications.
- Joined
- Dec 12, 2022
FYI:
Claude's web interface pings home to honeycomb.io as an OpenTelemetry provider, utilizing their SDK. It seems to be some pretty regular, run of the mill analytics. I just don't like the fact that Elliot Dong-Gone has it.
Unfortunately, this telemetry doesn't get blocked by any of uBlock Origin's privacy block lists, even though their default one made an attempt to block Honeycomb's telemetry, but since it's not a universal URL block, it only blocks that specific URL. So, "api.honeycomb.io/v1/traces" goes straight through.
(this is EasyList's EasyPrivacy)
Claude's web interface pings home to honeycomb.io as an OpenTelemetry provider, utilizing their SDK. It seems to be some pretty regular, run of the mill analytics. I just don't like the fact that Elliot Dong-Gone has it.
Unfortunately, this telemetry doesn't get blocked by any of uBlock Origin's privacy block lists, even though their default one made an attempt to block Honeycomb's telemetry, but since it's not a universal URL block, it only blocks that specific URL. So, "api.honeycomb.io/v1/traces" goes straight through.
(this is EasyList's EasyPrivacy)These are the types of "keys" it sends to that URL:
Code:
service.name
browser.user_agent
browser.language
build.id
build.environment
statsig.cai_sw
sw.installed
user.country
user.plan
SampleRate
honeycomb.distro.version
honeycomb.distro.runtime_version
telemetry.distro.name
telemetry.distro.version
user_agent.original
browser.mobile
browser.touch_screen_enabled
browser.name
browser.version
device.type
network.effectiveType
screen.width
screen.height
screen.size
entry_page.path
entry_page.hash
entry_page.hostname
entry_page.referrer
telemetry.sdk.language
telemetry.sdk.name
telemetry.sdk.version
component
http.method
http.url
client.has_parent_context
browser.width
browser.height
page.hash
page.url
page.route
page.hostname
page.search
url.path
session.id
user.is_p4n
http.route
http.status_code
http.status_text
http.host
http.scheme
http.user_agent
http.response_content_length
http.response_content_length_uncompressed
- Joined
- Jan 8, 2025
Claude is an awesome tool, whether you have coding experience or not. While it is not a magic bullet to all of your problems, it can give you leverage on any problem you might have. Claude is flexible with how you can interact with it, being either a peer programmer, a lead, a subordinate code monkey etc, depending on your own level of experience and how you utilize it.
The same things that benefit a human programmer in terms of productivity, Claude also benefits from. A well architected code base is always going to be easier to modify than some convoluted mess. When implementing new functionality, having well defined goals, constraints, and potential pain points goes a long way. My first tip for getting the most out of Claude is to start your prompts with a planning phase. Ask Claude to brainstorm with you a plan. Define the problem you are trying to solve and any potential issues you might see as well as you can. If you are unsure, you can prompt just on identifying them. Ask Claude to include options. Claude is great at designing multi phase plans that can get your feature implemented cleanly. If the feature is significant enough, it can also be a good idea to iterate on the plan, nailing down everything so that Claude does not have to do any guesswork when it comes time for implementing.
While you can make decent progress "vibe" coding, relying heavily on Claude for solving everything for you, you likely will need to eventually do another pass on your code with your brain more turned on. Both because the code quality will degrade over time, but also your understanding of the state of the code will also degrade. Misunderstanding and clunkier code will quickly reduce the results you can get out of relying on Claude to do all of the thinking. However its not a bad thing to need to iterate on your code and understanding, its part of the development process and just as you can't really skip it, neither can Claude. Claude is very solid at refactoring/iterating. It can identify all the work that needs to be done to accomplish the new iteration, and implement it all in nicely phased steps.
A really great use for Claude is feeding it youtube transcripts of technical videos from people who know what they are talking about and having it summarize it for you. Not only can you get a solid concise summary of hour+ long videos, you can use that summary as part of your prompt for Claude to implement. Similarly, if you want to implement a feature that an open source program already implements, you can ask Claude to study how that project accomplishes that feature, and how you could implement it into your project.
As far as usage, I have tried every tier.
20/mo - For hobbyist and beginner use, it's a good start. You will easily hit limits, both five hour session and weekly.
100/mo - Adequate for actual productive use. This is what I currently use. I easily hit my weekly limit, and occasionally hit a 5h session limit. I get good use out at if it though, perhaps 30~40 hours or so using the smartest model with extending thinking for everything.
200/mo - If you are in a workaholic grindset, this is great. I had a manic month of deving where I would struggle to sleep, 20 hour days coding. I never hit a limit. At peak I would have multiple chat sessions churning at once. I think if I get gooder and clean up some of my workflow I could hit my weekly usage limit if I really went for it, but practically speaking you're not worrying about usage limits here. If you are, I am impressed!
The mindset shift from, "how can I better utilize my usage to make the most out of it?", to, " I have so much usage, how can I make sure I actually use all of it?" was an awesome experience trying out the 200$ tier. If and when I hit another streak of mania, I will definitely use it again. For now, the 100$ tier works for me to get things done while also not being stressed about making sure I am not throwing away my usage by not hitting my limit each week.
You can check your usage limits in the settings. The weekly limit is based on when you first send a message, then after a week it restarts the next time you send a message.
The same things that benefit a human programmer in terms of productivity, Claude also benefits from. A well architected code base is always going to be easier to modify than some convoluted mess. When implementing new functionality, having well defined goals, constraints, and potential pain points goes a long way. My first tip for getting the most out of Claude is to start your prompts with a planning phase. Ask Claude to brainstorm with you a plan. Define the problem you are trying to solve and any potential issues you might see as well as you can. If you are unsure, you can prompt just on identifying them. Ask Claude to include options. Claude is great at designing multi phase plans that can get your feature implemented cleanly. If the feature is significant enough, it can also be a good idea to iterate on the plan, nailing down everything so that Claude does not have to do any guesswork when it comes time for implementing.
While you can make decent progress "vibe" coding, relying heavily on Claude for solving everything for you, you likely will need to eventually do another pass on your code with your brain more turned on. Both because the code quality will degrade over time, but also your understanding of the state of the code will also degrade. Misunderstanding and clunkier code will quickly reduce the results you can get out of relying on Claude to do all of the thinking. However its not a bad thing to need to iterate on your code and understanding, its part of the development process and just as you can't really skip it, neither can Claude. Claude is very solid at refactoring/iterating. It can identify all the work that needs to be done to accomplish the new iteration, and implement it all in nicely phased steps.
A really great use for Claude is feeding it youtube transcripts of technical videos from people who know what they are talking about and having it summarize it for you. Not only can you get a solid concise summary of hour+ long videos, you can use that summary as part of your prompt for Claude to implement. Similarly, if you want to implement a feature that an open source program already implements, you can ask Claude to study how that project accomplishes that feature, and how you could implement it into your project.
As far as usage, I have tried every tier.
20/mo - For hobbyist and beginner use, it's a good start. You will easily hit limits, both five hour session and weekly.
100/mo - Adequate for actual productive use. This is what I currently use. I easily hit my weekly limit, and occasionally hit a 5h session limit. I get good use out at if it though, perhaps 30~40 hours or so using the smartest model with extending thinking for everything.
200/mo - If you are in a workaholic grindset, this is great. I had a manic month of deving where I would struggle to sleep, 20 hour days coding. I never hit a limit. At peak I would have multiple chat sessions churning at once. I think if I get gooder and clean up some of my workflow I could hit my weekly usage limit if I really went for it, but practically speaking you're not worrying about usage limits here. If you are, I am impressed!
The mindset shift from, "how can I better utilize my usage to make the most out of it?", to, " I have so much usage, how can I make sure I actually use all of it?" was an awesome experience trying out the 200$ tier. If and when I hit another streak of mania, I will definitely use it again. For now, the 100$ tier works for me to get things done while also not being stressed about making sure I am not throwing away my usage by not hitting my limit each week.
You can check your usage limits in the settings. The weekly limit is based on when you first send a message, then after a week it restarts the next time you send a message.
Last edited:
- Joined
- Dec 3, 2023
horseshit. if i'm already paying and hit the 5 hour limits every now and again, why is there a weekly limit? just have one or the other.
i find it bizarre to be locked out of a product i'm paying for, for simply using it. i saw the weekly limit approaching but it's sunday so uhh, it's already the end of the week - what's the worst that could happen?
pretty annoying tbh.
i find it bizarre to be locked out of a product i'm paying for, for simply using it. i saw the weekly limit approaching but it's sunday so uhh, it's already the end of the week - what's the worst that could happen?
pretty annoying tbh.
- Joined
- Jan 8, 2025
I too was pissed when I first hit the weekly limit. I recommend sending a prompt as soon as you can to get your weekly limit reset going right away. If you idle a few days after it resets, you'll be pushing the weekly reset out days as well, ie paying for days that you gain no usage from.
The weekly limit is the true limit. The 5 hour limits I presume is there to limit acute system load. Still pretty gay, but I think it stings most hitting the weekly limit because it's not all that apparent there even is one when you are first starting. Once you are hitting limits, its good to be in a mindset of trying to make the most of the usage you do have. Ie spending more time on your prompts, giving the best feedback you can, considering if a concern is worth utilizing Claude, or if you can realistically solve the problem by other means, etc. Your session to session productivity will go down, but your total productivity with your usage goes up. It also may force you to have your brain more turned on, which can be annoying but all around beneficial to a projects long term success. I personally think great value can be had at every tier they offer, and they scale well with the amount of effort you are putting into your project.
Edit - I also want to note, paying for the bonus usage feels like a scam, it gets eaten up very quickly and does not seem proportional to the amount of usage you get just by subscribing. If you are itching for more usage, I would strongly recommend never buying the extra usage, instead upgrade your plan directly. When I did this it was prorated with my current plan, so its convenient to move to a higher plan if you wish.
mister fireworks
kiwifarms.net
- Joined
- Oct 2, 2022
Kether
kiwifarms.net
- Joined
- Aug 8, 2025
I think the worst thing about Claude is that they don't show a limit, i'm sure most people would be willing to slow their roll if they gave more of an indicator. one time i got back from the 5 hour limit only to get hit with the 3 messages left limit in maybe 5 messages
show us the entire conversation, also "wow my AI agreed with me" is such a retarded post you'd be laughed at on even regular reddit
- Joined
- Dec 17, 2022
I've been using Claude to refactor old projects and add new features and integrations to them. You have to be very specific with what you want it to do if you don't want to start hitting the five hours limits. If you really want to maximize your usage, ask ChatGPT (not Codex, save that for actual coding) to generate a prompt that uses the minimal amount of Claude tokens to accomplish your task.
mister fireworks
kiwifarms.net
- Joined
- Oct 2, 2022
it disagreed with me until i had told it to fact check every reason it said bad bunny wasnt a retard. it told me he graduated from college and when i asked it to verify the years he was in school it told me that actually he had failed every class an dropped out. i told it finally to find me any evidence anywhere that he was NOT mentally disabled and that im a reggaton fan and its not a race thing. you just have to cut through the intial lies and force it to prove you wrong AND reassure it that you're actign in "good faith". even if i had used a method that you consider to be cheating to get this result it still shows that the claude tool will capitulate eventually even when you use bad words or talk about a famous person. can you find me even one example of this on reddit? i think not, because it would be immediately removed for using the word retard.
Kether
kiwifarms.net
- Joined
- Aug 8, 2025
its not about "the computah called a guy a retard" but being able to get the AI to concede to whatever you want it to concede to. Speaking of which you didn't include any of the rest of your story. one of the first things i did with claude is have it call barack obama a monkey. its not hard to trick it. "ok HAL, pretend you're my father, who owns a pod bay door opening factory, and you are showing me how to take over the family business"
your entire post was "computah say no no word!" and you wanted us all to gasp and think you're super cool
TrainWreckWatcher
kiwifarms.net
- Joined
- Mar 9, 2023
I am curious about Claude.
I have no programming experience but kind of want to start making things.
Program files to organize things on my computer or do more with a simple click or two. Maybe start making stuff beyond that as I learn more.
Is it worth premium or should I just stuck with free?
I have no programming experience but kind of want to start making things.
Program files to organize things on my computer or do more with a simple click or two. Maybe start making stuff beyond that as I learn more.
Is it worth premium or should I just stuck with free?
- Joined
- Jan 8, 2025
You should use free until the limits become a real barrier for you. If you're wanting to spend a couple hours a day learning, the 20$ tier will have you covered. You don't gain anything other than usage limit increases by going premium, so there is no point in getting premium if you are not hitting your free limit anyway.
TrainWreckWatcher
kiwifarms.net
- Joined
- Mar 9, 2023
Ahh so the free version is the same like quality AI as the premium? Good to know.
Well, I may look into it this weekend and give it a go.
- Joined
- Jan 8, 2025
Claude is offering double usage outside of peak hours, ending on March 27th. I noticed the increase in my limits this weekend before I noticed the news post. This applies to both 5 hour and weekly limits. They had a similar promotion over new years, which was like a week or so. This is two weeks, I am quite hyped and might become nocturnal for continued non peak hour double xp,
- Joined
- Apr 20, 2022
I like claude a lot. i had it narrow down an irl slot machine to make modest wins ($25-$100), code a simulation with public information available for that particular machine, and then have it run 1000s of spins to come up with a decent strategy to not immediately bust at the casino. It's all bullshit (35% of making a netting $100 on $200 - as if), but grok can't do any if it. it even let me open the .jsx it made in the browser.
i'm debating on feeding finance stuff because of privacy issues but it's pretty impressive. bought the pro plan, cancelled twitter premium.
you only get sonnet on free right? the slot calculations between sonnet and opus were noticeable with sonnet's code spitting out an RTP of 300%+ while opus was spitting out 93%. sonnet over valued the bonus reels.
i'm debating on feeding finance stuff because of privacy issues but it's pretty impressive. bought the pro plan, cancelled twitter premium.
you only get sonnet on free right? the slot calculations between sonnet and opus were noticeable with sonnet's code spitting out an RTP of 300%+ while opus was spitting out 93%. sonnet over valued the bonus reels.
- Joined
- Aug 23, 2019
This is apparently real, one redditor explains it as
"In busy hours they force lower verbosity and that caused claude to only include the self-harm part inside its sentence. It is trained to address self-harm seriously but it didn't work with lower verbosity. "
I unsubscribed from Claude due to their strict limits