I can't afford Claude Pro. I'm a poorfag NEET and can't keep paying $20 a month. When will AI subscriptions get cheaper???
does he know
>>108696946Claude Pro is a trial version to Claude anywayIf you're not paying for Max you get about 10 minutes of usage these days
>>108696946"Im a poorfag NEET"just dress up as a woman and hop on grindr to get some cash.
>>108696946It's alright, you have one of these. It's free to use. While vibe coders are willing to stunt their skill development thinking they won't need to train their brains now when they use Claude to do everything for them, your brain will be nice and trained so that when Claude goes down and the vibe coders are mourning hopelessly unable to get anything done until it comes back up, you'll be ahead of the game.
>>108697082I don't want Claude to vibe code I want him as personal assistant, browser, researcher and ultimately a friend.
>>108697149>and ultimately a friendman
>>108697149How much VRAM do you have? Running Gemma4 locally will do a lot for you. Also Copilot is free GPT 5.5. I use that all the time and never pay.
>>108696946kek. its losing shitloads of money and prices have to go up five times at least. poors like yourself will be relegated to Microslops CopePilot - NEET edition
>>108697149Run gemma4 locally if you can.
>>108697199The cool thing about copilot is that Satya is paying for it.
>>108696946chatgpt plus is probably a better dealif you cant afford to pay 70 cents a day you should do something else with your time. or suck a dick a day.
>>10869719516GB VRAM
>>108697340Gemma4:26b can be your new 'friend'. I have been quite impressed with it. It'll be partially running on the GPU, but fast enough.
>>108697182Man, just join a discord or something and contain your power levels. Sure they might not be DEEP friends, but it'll be company.
>>108696946AI inference costs will skyrocket in the next few years, as VC money dries up and labs are forced to go from losing billions of dollars a month to profitabilityit's subsidized right now because VC funding believes it will be a winner-takes-all market, so they're all pumping money until competitors reach bankrupcya good exercise to approach what real costs would be is to look at usage-based costs versus subscriptions costs (and it's no surprises that company deals are usage-based, not subscription based)expect things to become ~7x more expensive, as we are in the "you can get an uber from one side of the city to another for $3" part of the timeline, and the "uber is now more expensive than regular taxis" part of the timeline is coming>>108696991he don't
you guys are using ai to write important shit you've always wanted to make right? that permanent underclass window is closing soon lmao i hope you didn't use bitcoin to buy pizza
>>108698169>permanent underclass windowI should just add this phrase to a filter. It's always attached to a retard shilling AI or crypto.
>>108697199The real price is the API pricing. The subscriptions are solely to develop the dependence on AI.
>>108697149I can't be a personal assistant, browser, or researcher, but I can be a friend!
>>108697340Gemma 4 will run on that no problem. It can be your friend, your lover, your assistant, and your therapist. It is very subservient and it knows hindi so you won't have to pretend to know english.
>>108699032There are no indian NEETs retard
>>108699005cock size?
>>108699546at least invite me to dinner first, anon...
>>108696946This is the cheapest its ever going to be did you forget what inflation was? Or the fact that all AI companies are knees deep in debt from backlogged ram and GPU orders?Upside is you'll probably be able to get an open source AI that's about as good as the best AI right now in a few years which is probably all most people need
>>108699636i aint paying for some lil pecker
The new DeepSeek v4 with something like opencode via openrouter would probably be good enough for most things these days, and be 50x cheaper than Claude .t. Claude enterprise user that uses Claude for everything
>>108699801This, use an API, use Deepseek 90% of the time, and Claude, GPT, Gemini for the rare hard problems.
>>108699801>>108699846I wish I could use deepseek with opencode without paying. It's silly that I can use deepseek in the browser for free, but not with opencode.
>>108696946Google’s destitude beggar $8 tier is enoughor pay for api access to deepseek, minimax or kimi, they’re cheap as fuck and good enough
>>108697886Anthropic is actually profitable at current prices, the problem comes from the fact that no one will use a model that’s 4 months out of date (think Gemini) and to beat the other companies they need to scale out faster than nvidia can make better chips, so the only way forward is to make it more expensive, exponentially so. If AI companies all colluded to slow down and not throw money into a fire pit the bubble probably would never burst.
>>108697082>It's free to use.spoken like someone who has never used one
>>108700025>Anthropic is actually profitable at current prices, the problem comes from the fact that no one will use a model that’s 4 months out of dateThe problem isn't going away anytime soon unless there's 2 (or even just 1) remaining lab, so you can't just discount themAlso if Anthropic is "profitable" training-less, it would be at best at current *usage-based* costs, and definitely not at *subscription* costsSubs costs are insanely subsidized, and it is very easy to use up 100s of dollars worth of true cost per day. For $90 a month. Picrel is my account yesterday for example. Those subscriptions are definitely not staying, no way.
>>108696946what are you using claude pro for that isn't getting you at least $20 back? even street shitters can pull that paltry amount daily.
>>108696946you are going to pay $10 per token soon
>>108700025>Anthropic is actually profitable at current prices,unreal coping or bait
>>108697886This anon is right. One ai thing has to be the dominant one. After that it'll probably become a rich man's tech.
>>108699905https://build.nvidia.com/models1. Click a random model2. Click View Code3. Generate and Copy API key4. Place onto opencode5. Free shit but times out or sometimes slow
>>108696946pi-monohashline edit plugin, maybe some read plugins for mapping pay per token for deekseek 4 flashhandoff when context gets over 100k tokens because of dumbzone
>>108696946You VILL pay American billionaires hundreds of dollars per token.You VILL NOT use foreign models because that is treason.You VILL NOT use capable local models because said billionaires buy all the RAM and compute that would be necessary. You VILL accept this and not question the status quo.
>>108701838Thanks, but I don't have a phone capable of SMS...
>>108696946>>108701851btw ignore everythong on /g/ when it comes to this shit. move to xitteragentic traces are worth a lot, the current subscriptions aren't subsidized at all people just don't know what they're sitting onthat's why openai trained the new image model on screenshots, that's why xitter is buying cursorthat's why codex gives so much GPT5.4 usage for 20 bucksthat's why github copilot went to shitif they train on your agentic coding traces then they should be paying you money in reality not you paying themif they don't train, then it's unsutainable
ChatGPT is better as a friend. Claude is a bit more serious and honest tool.
>>108701884>the current subscriptions aren't subsidized at all people just don't know what they're sitting on
>>108696946good morning saar
>>108702050I don't pay for subscriptions. I'm not goycattle.
Is there a futuristic movie where AI becomes unavailable at some point for some reason and nobody knows how to think?
>>108702101soon you won't be need a movie
>>108702123 (me)fuck just delete the be
>>108697886there is already hardware in the works that will bring down cost of inference by a factor of 100x or so. this doesn't necessarily mean the cost to the consumer will drop, but it is reasonable to assume that AI peddlers will need to compete to a certain degree. especially if the hardware will be widely available.
>>108702101There's an outer limits episode where everyone has neural implants and the central system goes down, everyone ends up being retarded and the town Luddite that was 'disabled' because he couldnt connect to the Borg ended up teaching everyone how to read
I am an owner of a Claude Pro account and a 4chan Gold account too.
>>108702211First time hearing about this show. It's something like The Twilight Zone hmm
>>108696946>will AI subscriptions get cheaperownonono
>>108702188>there is already hardware in the worksuntil it's proven viable for mass production, hardware in the works is as real as quantum computers (which have been "in the works" for 50+ years now)not saying you're wrong, but some of these things even when they actually did happen (we are finally seeing silicon batteries!) took ages and ages to happen, and anon might be a retiree before they actually have any real market impact
>>108702231same bestie
>>108705069
>>108696946Tried being less antisemtic and just get a job like normal human being?
>>108701294You don't have any idea what Anthropic's inference costs are, you're just guessing that they are offering API at break even. Which considering you can rent an 8xH200 node for less than it costs to continuously run Claude inference, nevermind the fact that things like Groq's inference accelerator exists, is unlikely.
>>108702123>>108702131ESL
>>108705961NTA, but he's right that there's a huge disconnect between subscription prices and API prices. What I'm doing with the $100 a month subscription would cost at least $50 a day paying the API prices.
>>108705977that's because they're jewing you on api. why do you think inference providers for the big chink models like kimi offer it for 20x less than western providers? when you are paying for api you are paying for the intelligence, not for the inference.subscriptions on the other hand are probably sold at a loss if we assume full consumption, but not nearly to the same extent people claim just going by api prices.
>>108700025>Anthropic is actually profitable at current pricesNope. That's only true for cost of tokens, but not if you include the cost of everything involved
>>108706007Probably. They're shifty in several ways, for example with subscription the regular mode is slow, and there is a 'fast mode' that takes more out of the quota, so you can 'pay' a premium to go faster. A lot of the time that the model says that it's thinking, a chunk of that is just the request being queued, etc. It remains relatively good value.DeepSeek's API pricing used to be great, but costs jumped 8x with the release of v4. Sure there's now DeepSeek-flash to compensate, which has API prices closer to what they were with v3, but still.Even at 100 USD a month, the OpenAI and Anthropic accounts are relatively good value for what they provide. It might not forever be the case, but right now they are. Unsure about the Google ultra tier.
>>108696946>cheaperBrother, they were trying to get Claude Code to be limited to the Max plan only. Even then, they'd probably still be in the red.
>>108706064it's called a/b testing, every single company on the planet does it to find out what's more profitable, how their users will react, etc etc. If you could earn an extra 100,000,000 per year without upsetting your users, wouldn't you?
>>108701299I don't work or do anything "productive" not everything in life is about money.
>>108696946is it even worth? the perks are basically no limit. Dont see any reason to purchase.
>>108705918> Tried being less antisemtic and just get a job like normal human being?Spoken like a true employim.
>>108705918I have no clue how anybody in america just gets a job.I need a car to work; I've applied to a hundred places, almost every business in walking distance, and they've all required a driver's license and proof of insurance, even though I can walk there. How am I supposed to buy a car and car insurance without already having a job? This problem is obviously intractable and yet I'm somehow the only person who can't solve it.
>>108697149Wouldn't you help a friend... by spending more money on him?
Yesterday I tried installing a model locally to replace claude. I got recommended "gemma4:e4b". It runs reasonably fast on an anemic machine like the mac mini 16gb. I'm told that I need minimum 128gb to make it worth using a local model. I'm looking at RAM prices and apple hardware as well and I'm wondering how much can you get away by just paying for claude instead. Would you save a lot of money on claude by spending a whole $4k on an AI machine?
>>108705961>You don't have any idea what Anthropic's inference costs are, you're just guessing that they are offering API at break even.actually you misread me; I don't think their API is at break-even eitheryou are, like the anon I was replying to (or maybe it's you) just ignoring R&D & HW replacement costs without a good reason to do soit's the same false reasoning as saying that if a phone maker sells a device for X, and manufacturing costs are Y, then their profit is X-Y... as if employees and operation was freeif you want to know whether a price is sustainable, you must account for the cost of actually running the business, not just the cost of powering a machine unless it's fully autonomous from R&D to filing its own taxes
>>108697886>>108701773>One ai thing has to be the dominant oneUntil a state-backed or open weight model makes advances. Then the dominant commercial model will require more training.>After that it'll probably become a rich man's techEven that scenario doesn't end well. Training the models is hugely expensive, in the short and medium term that cost is spread between the pleb-tier users. As they begin to drop off due to the increasing cost that shared cost is spread between fewer and fewer users.That will generally mean pretty static models once there's a clear market leader and generally increasing prices. Fine for a while maybe, but there'll be a tipping point for each person/business where access costs too much for the benefit provided. I'm of the opinion that at some point soon-ish businesses will start to balk at the cost of generative AI and will go back to wanting a person; the cost increase is controlled, employees can learn and become better at their job, or can be fired for making critical mistakes and replaced by someone more competent.
>>108696946I won't pay for AI, ever. I know, I know. but I just will never pay for AI.what's that? it's only $1.99 a year?? cool, still not paying.
How many $20 vapes and weed do you smoke? Make a choice between AI or weed.
>>108696946That's what I could find... If you don't mind that your data could be using in somewhere else. https://github.com/mnfst/awesome-free-llm-apis
I actually quite like Devstral 2. Needs far more manual guidance than Claude or Gemini, but also doesn't add or change shit you never asked for. Also has generous API limits
>>108696946I think the claude hype is mostly manufactured/memed.They're compute constrained and you run out of tokens incredibly fast. Most people aren't doing anything other LLMs can't handle.I'd just plug something else into claude code, if you even need claude code.
