Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Shells Virtual Desktop
BMail.ag - Secure Email Service
Server.net
CPLicense.net
VPS Server
Buy VPN
Vultr
VMs for AI
HostDare
ReliableSite White-Label Dedicated Hosting for Resellers
InterServer VPS
BMail.ag - Secure Email Service
Best VPN
High-Performance Bare Metal Server Solutions
Karvl.com
Server Mania Cloud Hosting
DataWagon Hosting
AlphaVPS Hosting
Evoxt.com
Clouvider
VPS Hosting with NVMe
Residential IPs in the US & 4G Mobile Proxies in EU & US with Unlimited Bandwidth
ReliableSite White-Label Dedicated Hosting for Resellers
Rabisu - Hosting Solutions
Shells Virtual Desktop
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Which AI tool have you been using the most lately?

13

Comments

  • rpqurpqu Member
    edited February 28

    @Kodomu said:

    @rpqu said:

    @Kodomu said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Most of the free tiers are so limited that it is not even worth the time. And this is ignoring that using the free tier is still assisting with the training and development of AI.
    If they were willing to pay me to sit and baby their model while providing feedback for its suggestions, then it would be different. It also increases their need for more server resources, even if small, so again works against me.

    That's why puppeteering matters. It may costs you few dollars to purchase $0.99 numeric .xyz, then purchase cheap voip (or sim if possible), then register at as much as AI platform if possible (which require phone #). Repeat few times if necessary. Then, write a puppeteer to automate things like, please make a summary of this reddit thread. Then ask explanation of controversial comment given previous conclusion... It's going to cost lots of token. Or, just fed it random government text and ask it to translate to other language. It definitely will cost them more than monthly voip fee...

    It's a lot more effort than it is worth just to use something that has largely produced garbage from the few times I have used it.

    That garbage came from clusters of $30k GPU. Which will have less than half of value for producing garbage in 2y, 3y tops.

    @emgh said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Where did you get 10 % utilization being a realistic figure from?

    I don't fully understand what you said. But, let me explain how that figure may came. It's revenue/(capex+opex+depreciation of existing capacity–remaining value of hardware).
    Last year, it's 40-60/(200B+50B+65B-100B)=18.66~27.9%
    In 2026, it's going to be ??B/(500B+60B+107.5B-250B)

  • stable_geniusstable_genius Member
    edited February 28

    @itachikonoha said:

    @stable_genius said:

    @itachikonoha said:

    @stable_genius said:

    Which AI agents have been using you as a tool lately?

    You think an AI agent is the tool when in fact you are the tool. Those agents are being trained on your input, you are the tool helping them improve.

    Remember that brief period of time when everyone worked for big tech, for free, by solving captchas? Well now we are working for them again, for free or even paying them, by training their AI.

    So the real question is:

    Which AI agents have been using you as a tool lately?

    Wrong in the sense that in your context, you are the subject (whom AI will train itself upon) and NOT the tool but AI is still the tool.

    AI is a tool that uses you as a tool. You are both tools.

    By that logic, every entity in universe is a tool because there is minimum a 2 way relationship when you involved with someone/something. Based upon which spectrum you stand on, each will be a tool.

    Nope. > @itachikonoha said:

    @stable_genius said:

    @itachikonoha said:

    @stable_genius said:

    Which AI agents have been using you as a tool lately?

    You think an AI agent is the tool when in fact you are the tool. Those agents are being trained on your input, you are the tool helping them improve.

    Remember that brief period of time when everyone worked for big tech, for free, by solving captchas? Well now we are working for them again, for free or even paying them, by training their AI.

    So the real question is:

    Which AI agents have been using you as a tool lately?

    Wrong in the sense that in your context, you are the subject (whom AI will train itself upon) and NOT the tool but AI is still the tool.

    AI is a tool that uses you as a tool. You are both tools.

    By that logic, every entity in universe is a tool because there is minimum a 2 way relationship when you involved with someone/something. Based upon which spectrum you stand on, each will be a tool.

    Nope, the relation is transitive not 2 way.

    The owner of the AI uses it as a tool to turn you into a tool, you and AI are not tools to each other you are both tools working for the owner of the AI. Transitive relation not 2 way, this is a very asymmetric relationship.

    It is an arrangement that never works to your benefit, you always take the short straw.

    Thanked by 1Kodomu
  • emghemgh Member, Megathread Squad

    @rpqu said:

    @Kodomu said:

    @rpqu said:

    @Kodomu said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Most of the free tiers are so limited that it is not even worth the time. And this is ignoring that using the free tier is still assisting with the training and development of AI.
    If they were willing to pay me to sit and baby their model while providing feedback for its suggestions, then it would be different. It also increases their need for more server resources, even if small, so again works against me.

    That's why puppeteering matters. It may costs you few dollars to purchase $0.99 numeric .xyz, then purchase cheap voip (or sim if possible), then register at as much as AI platform if possible (which require phone #). Repeat few times if necessary. Then, write a puppeteer to automate things like, please make a summary of this reddit thread. Then ask explanation of controversial comment given previous conclusion... It's going to cost lots of token. Or, just fed it random government text and ask it to translate to other language. It definitely will cost them more than monthly voip fee...

    It's a lot more effort than it is worth just to use something that has largely produced garbage from the few times I have used it.

    That garbage came from clusters of $30k GPU. Which will have less than half of value for producing garbage in 2y, 3y tops.

    @emgh said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Where did you get 10 % utilization being a realistic figure from?

    I don't fully understand what you said. But, let me explain how that figure may came. It's revenue/(capex+opex+depreciation of existing capacity–remaining value of hardware).
    Last year, it's 40-60/(200B+50B+65B-100B)=18.66~27.9%
    In 2026, it's going to be ??B/(500B+60B+107.5B-250B)

    This is some cost recovery ratio-ish type of calculation. It does not measure utilization at all.

    You could have 90% utilization and still have a low cost recovery ratio if your pricing is too low, or you could have 30% utilization but charge enough to cover everything.

  • stable_geniusstable_genius Member
    edited February 28

    @Kodomu said:

    @itachikonoha said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Adaptation is something that one must implement at times in my opinion. In order to fight against the enemy, you need to survive first.

    I don't need to use AI at all so there is nothing to survive against, the things I could use it for would save me very minimal time if any at all.

    The enemy wins if I start using AI instead of learning how to do things myself.

    AI agents are being deployed with the purpose of deskilling independent programmers so that at as soon as possible only the owners of AI agents and no one else will be able to produce software.

    Their intent is obvious they want to kill competition and they hope to achieve that by deskilling programmers while using them to upskill the AI agents. All that while charging their "users" for the "benefit" of being thrown under the bus.

    Thanked by 1Kodomu
  • emghemgh Member, Megathread Squad

    @stable_genius said:

    @Kodomu said:

    @itachikonoha said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Adaptation is something that one must implement at times in my opinion. In order to fight against the enemy, you need to survive first.

    I don't need to use AI at all so there is nothing to survive against, the things I could use it for would save me very minimal time if any at all.

    The enemy wins if I start using AI instead of learning how to do things myself.

    AI agents are being deployed with the purpose of deskilling independent programmers so that at as soon as possible only the owners of AI agents and no one else will be able to produce software.

    Their intent is obvious they want to kill competition and they are doing that by deskilling programmers.

    wow i've never thought about it like that, that's so incredibly true

    Thanked by 1cainyxues
  • @stable_genius said:

    @Kodomu said:

    @itachikonoha said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Adaptation is something that one must implement at times in my opinion. In order to fight against the enemy, you need to survive first.

    I don't need to use AI at all so there is nothing to survive against, the things I could use it for would save me very minimal time if any at all.

    The enemy wins if I start using AI instead of learning how to do things myself.

    AI agents are being deployed with the purpose of deskilling independent programmers so that at as soon as possible only the owners of AI agents and no one else will be able to produce software.

    Their intent is obvious they want to kill competition and they hope to achieve that by deskilling programmers while using them to upskill the AI agents. All that while charging their "users" for the "benefit" of being thrown under the bus.

    Everyone’s worried about AI replacing programmers, but as NDR systems become more powerful and AI-driven, what safeguards exist to prevent them from evolving into large-scale surveillance infrastructure; especially if adopted at ISP or government levels? Sure they monitor people now via cameras and so on. What happens if your active connection is already being read by the AI since your ISP is running it.

  • @NushairAlvi said:

    @sunriseing said:
    use mostly Gemini and Claude code with z.ai as Provider

    Copilot is better than any other AI in the field of Free AI. I always prefer to use VS Code for coding. But Gemini is also good for Study.

    i use gemini bc i share the premium with friends (3,50 € for gemini premium and 2tb), vs code i also use but i use it with claude code but with glm 5 from z.ai

    Thanked by 1NushairAlvi
  • rpqurpqu Member
    edited March 1

    @emgh said:

    @rpqu said:

    @Kodomu said:

    @rpqu said:

    @Kodomu said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Most of the free tiers are so limited that it is not even worth the time. And this is ignoring that using the free tier is still assisting with the training and development of AI.
    If they were willing to pay me to sit and baby their model while providing feedback for its suggestions, then it would be different. It also increases their need for more server resources, even if small, so again works against me.

    That's why puppeteering matters. It may costs you few dollars to purchase $0.99 numeric .xyz, then purchase cheap voip (or sim if possible), then register at as much as AI platform if possible (which require phone #). Repeat few times if necessary. Then, write a puppeteer to automate things like, please make a summary of this reddit thread. Then ask explanation of controversial comment given previous conclusion... It's going to cost lots of token. Or, just fed it random government text and ask it to translate to other language. It definitely will cost them more than monthly voip fee...

    It's a lot more effort than it is worth just to use something that has largely produced garbage from the few times I have used it.

    That garbage came from clusters of $30k GPU. Which will have less than half of value for producing garbage in 2y, 3y tops.

    @emgh said:

    @rpqu said:

    @Kodomu said:
    None, I don't wish to use or fund something that works against me.
    I'd like to not pay 3-5x the price for RAM and storage again.

    Just use free tier. It would just increase the potential revenue and add opex.
    The day of reckoning will come when they realize that building excess capacity doesn't make sense. It's impossible to absorb capex if the utilization rate is below 25%, 20%, or even 10%. Their customers would demand discount if there's a newer GPU with more VRAM, which means less I/O wait, despite bigger opex

    Where did you get 10 % utilization being a realistic figure from?

    I don't fully understand what you said. But, let me explain how that figure may came. It's revenue/(capex+opex+depreciation of existing capacity–remaining value of hardware).
    Last year, it's 40-60/(200B+50B+65B-100B)=18.66~27.9%
    In 2026, it's going to be ??B/(500B+60B+107.5B-250B)

    This is some cost recovery ratio-ish type of calculation. It does not measure utilization at all.

    You could have 90% utilization and still have a low cost recovery ratio if your pricing is too low, or you could have 30% utilization but charge enough to cover everything.

    It's impossible to measure the real utilization rate. Thus, we could approximate the utilization rate based on payment from the industry against the capex, depreciation and leftover value from last year balance sheet. Of course, this is based on the assumption that hyperscaler aren't a summerhost that sells at cost or even lower than the actual cost. So, the real utilization must be lower than this approximation.

    Note: opex removed because we're trying to approximate utilization rate, not money earned: money spent

  • @Nanja said:

    @Patriarch said:
    Antigravity with Gemini and Claude
    OpenCode with GLM and Kimi

    I’ve never used Antigravity. It looks like a neat IDE, and I’m curious. Why do you use Gemini + Claude instead of just one or the other?

    Yeah it's kinda neat and have enough features for me
    For claude + gemini it's like switching due to not enough quotas lol
    When Opus quota run out, I switch using latest Gemini to complete the rest of codes haha

  • @suyadi92 said:

    @Nanja said:

    @Patriarch said:
    Antigravity with Gemini and Claude
    OpenCode with GLM and Kimi

    I'm curious too about antigravity, been using z.ai and copilot (copilot is better bang for the bucks imo). In your experience of using antigravity, is it cheaper/better than Copilot for coding needs @Patriarch ?

    Can't say anything about it as I never try copilot before haha
    I tried antigravity with Google AI Pro trial for the 1st time before and it suited my needs
    I'll try copilot soon

    Thanked by 1suyadi92
  • Claude. Loving claude code, tho honestly mostly for the lazy crap that doesn't really matter that I can't be bothered to spend time on myself, lol. Kept burning the quota for pro, eventually swapped to max. Expensive, but it's nice not having to worry about quota.

    Yesterday/Tonight's random claude code session: minimalistic dark mode userscript for LET that includes a nice module at the top to stalk all of @host_c posts :joy:

    Thanked by 1COLBYLICIOUS
  • Interesting to see that there are not too many here using Codex for coding. Found that it's been more dependable and "trustworthy", if you can call it that, over Gemini and Claude. The latter two are definitely incredibly intelligent and capable, but I think can be a bit lazy and less of a "senior engineer" than I'd like. Codex feels better to me in that regard, with GPT-5.4.

  • rpqurpqu Member
    edited April 1

    Resurrecting this thread over Claude's crazy-low limit on non-API. It feels the limit within the last few days are lower than 2 weeks ago.

  • aphexaphex Member

    @rpqu said: Resurrecting this thread over Claude's crazy-low limit on non-API. It seems I'm the limit within the last 10 days are lower than 2 weeks ago.

    The last 2 weeks was a 2x promo.

    How it works: - 2x usage on weekdays outside 5–11am PT / 12–6pm GMT - 2x usage all day on weekends - Automatic, nothing to enable

    This ended few days ago. So you are halved now.

    ADDITIONALL for non-api: https://www.theregister.com/2026/03/26/anthropic_tweaks_usage_limits/ (In addition to halved quota)

    So you are more than Halved now during weekend and off peak.

  • bbn12bbn12 Member

    @rpqu said:
    Resurrecting this thread over *** crazy-low limit on non-API. It seems I'm the limit within the last 10 days are lower than 2 weeks ago.

    agree, if they keep this going. they are done in a private-customer arena... unless u go with API expensive mode

  • rpqurpqu Member

    @aphex said:

    @rpqu said: Resurrecting this thread over Claude's crazy-low limit on non-API. It seems I'm the limit within the last 10 days are lower than 2 weeks ago.

    The last 2 weeks was a 2x promo.

    How it works: - 2x usage on weekdays outside 5–11am PT / 12–6pm GMT - 2x usage all day on weekends - Automatic, nothing to enable

    This ended few days ago. So you are halved now.

    ADDITIONALL for non-api: https://www.theregister.com/2026/03/26/anthropic_tweaks_usage_limits/ (In addition to halved quota)

    So you are more than Halved now during weekend and off peak.

    I'm using it outside of the peak hour. And it feels like it's like I'm receiving quarter the quota, despite it was run on non-peak hour.
    Anyway, I saw something
    https://old.reddit.com/r/ClaudeAI/comments/1s8zxt4/thanks_to_the_leaked_source_code_for_claude_code/

    Thanks to the leaked source code for Claude Code, I used Codex to find and patch the root cause of the insane token drain in Claude Code and patched it. Usage limits are back to normal for me!
    https://github.com/Rangizingo/cc-cache-fix/tree/main

    Disclaimer : Codex found and fixed this, not me. I work in IT and know how to ask the right questions, but it did the work. Giving you this as is cause it's been steady for the last 2 hours for me. My 5 hour usage is at 6% which is normal! Let's be real you're probably just gonna tell claude to clone this repo, and apply it so here is the repo lol. I main Linux but I had codex write stuff that should work across OS. Works on my Mac too.

    Also Codex wrote everything below this, not me. I spent a full session reverse-engineering the minified cli.js and found two bugs that silently nuke prompt caching on resumed sessions.

    What's actually happening Claude Code has a function called db8 that filters what gets saved to your session files (the JSONL files in ~/.claude/projects/). For non-Anthropic users, it strips out ALL attachment-type messages. Sounds harmless, except some of those attachments are deferred_tools_delta records that track which tools have already been announced to the model.

    When you resume a session, Claude Code scans your message history to figure out "what tools did I already tell the model about?" But because db8 nuked those records from the session file, it finds nothing. So it re-announces every single deferred tool from scratch. Every. Single. Resume.

    This breaks the cache prefix in three ways:

    The system reminders that were at messages[0] in the fresh session now land at messages[N] The billing hash (computed from your first user message) changes because the first message content is different The cache_control breakpoint shifts because the message array is a different length Net result: your entire conversation gets rebuilt as cache_creation tokens instead of hitting cache_read. The longer the conversation, the worse it gets.

    The numbers from my actual session Stock claude, same conversation, watching the cache ratio drop with every turn:

    Turn 1: cache_read: 15,451 cache_creation: 7,473 ratio: 67% Turn 5: cache_read: 15,451 cache_creation: 16,881 ratio: 48% Turn 10: cache_read: 15,451 cache_creation: 35,006 ratio: 31% Turn 15: cache_read: 15,451 cache_creation: 42,970 ratio: 26% cache_read NEVER moved. Stuck at 15,451 (just the system prompt). Everything else was full-price token processing.

    After applying the patch:

    Turn 1 (resume): cache_read: 7,208 cache_creation: 49,748 ratio: 13% (structural reset, expected) Turn 2: cache_read: 56,956 cache_creation: 728 ratio: 99% Turn 3: cache_read: 57,684 cache_creation: 611 ratio: 99% 26% to 99%. That's the difference.

    There's also a second bug The standalone binary (the one installed at ~/.local/share/claude/) uses a custom Bun fork that rewrites a sentinel value cch=00000 in every outgoing API request. If your conversation happens to contain that string, it breaks the cache prefix. Running via Node.js (node cli.js) instead of the binary eliminates this entirely.

    Related issues: anthropics/claude-code#40524 and anthropics/claude-code#34629

    I don't notice this because I turned off the auto-update

  • ElliotJElliotJ Member

    @rpqu said: Resurrecting this thread over Claude's crazy-low limit on non-API. It feels the limit within the last few days are lower than 2 weeks ago.

    Currently using a combination of OpenAI (via OAuth, Plus plan), Alibaba's Coding Plan, and Z.ai coding plan (paid $20ish for a year of lite) - In that order.

    Gipperty for orchestration, Qwen/GLM for execution, GLM via Z.ai in the unlikely chance I run out of usage.

    We're in a weird timeframe where coding plans are popping up all over the place, might be worth trying out one or two to see if they work for you. Admittedly, I haven't used Anthropic's models directly so I don't have a comparison, but it's working alright for me. I'm not employed for dev work so YMMV.

  • dosaidosai Member

    Mostly Gemini because I have free pro. Thinking about buying claude pro.

  • aphexaphex Member

    @ElliotJ said: Gipperty for orchestration, Qwen/GLM for execution, GLM via Z.ai in the unlikely chance I run out of usage.

    Do you happen to know what payment methods zai takes or what merchant/company they bill as? It look interesting but I have to pay heavy intl fees transaction

  • woinokizwoinokiz Member

    I just rebuilt leaked Claude code source , now it's gonna be my main tool after I add support for certain api types

  • NushairAlviNushairAlvi 🚩 Host Rep Tag Suspended

    Using Gemini for study

  • MonocleMonocle Member

    @forest said:
    I use DeepSeek very occasionally if I forget a word or term. That's pretty much all.

    You can google search the word and get a very quick definition/meaning. I just select the word, right click and thensearch google

  • ElliotJElliotJ Member

    @aphex said: Do you happen to know what payment methods zai takes or what merchant/company they bill as? It look interesting but I have to pay heavy intl fees transaction

    Honestly, I can't remember - I have cards with 0% international fees, so it wasn't something on my conscience.

    Z.ai is in the last place for a reason though, when I signed up with them in Jan (I paid for a year, apparently 3x Claude Pro usage, for $28) it was stupidly slow and unreliable. I'm not sure if that's improved, but I'd tread lightly for up to date reviews.

  • suyadi92suyadi92 Member
    edited April 1

    @aphex said:

    @ElliotJ said: Gipperty for orchestration, Qwen/GLM for execution, GLM via Z.ai in the unlikely chance I run out of usage.

    Do you happen to know what payment methods zai takes or what merchant/company they bill as? It look interesting but I have to pay heavy intl fees transaction

    They use Orb Billing if I'm not mistaken. Tried it once but tokens allocation was super tight made me switched back to copilot. I still have the invoice, let me check

    Thanked by 2rpqu aphex
  • ThrowRaPestThrowRaPest Member, Patron Provider

    @LBF said: I am very sorry, didn't mean to insult someone. Won't do that again.

    My apologies, I made a mistake.

    What are we talking about? or who are we talking about?

  • GPT-5.4 is the best for now. Hope Codex CLI could learn something from the source of Claude Code.

  • eva2000eva2000 Veteran

    Still Claude Code on Max $100 plan paired with ChatGPT Plus $20, Gemini AI Pro $20 and GLM Coding Pro plan $129/yr discounted and Github Copilot Pro free via open source maintainer program :smiley:

    Claude Code is primary used 5.5 billion tokens the past 30 days of which 3.3 billion tokens during recent 2x off-peak usage 2 week promo.

    Added my Claude Code ai-image-creator skill so Claude Code can create images and leverages Google Gemini Nano Banana 2, Flux 2, Riverflow V2 Pro, Bytedance Seedream 4.5, OpenAI GPT-5 Image LLM models
    https://github.com/centminmod/my-claude-code-setup

    Examples

    https://www.threads.com/@george_sl_liu/post/DWj1-gWk1il?xmt=AQF0K2lQnUhz5abjz34ijzkS8-wIw0fBlhob2olL2sJuikRqUe9BeqXeuDX0wsvZbCx79KM&slof=1

    https://www.threads.com/@george_sl_liu/post/DWjfrpJk5_D?xmt=AQF0mv0sm-aJ7k5ha-oRXJ-QUNZ0F2oJbxPxpBiXJY9KAoMoufhMsJmmeR0wftUpsQftjj4&slof=1

    Thanked by 1abtdw
  • rpqurpqu Member

    @eva2000 said:
    Claude Code is primary used 5.5 billion tokens the past 30 days of which 3.3 billion tokens during recent 2x off-peak usage 2 week promo.

    The date is April 1 2026, not early 2027. So, I'm sure 5.5B tokens would have costs more than $6k. Are you paying out of your own pocket or not?

  • eva2000eva2000 Veteran
    edited April 1

    @rpqu said:

    @eva2000 said:
    Claude Code is primary used 5.5 billion tokens the past 30 days of which 3.3 billion tokens during recent 2x off-peak usage 2 week promo.

    The date is April 1 2026, not early 2027. So, I'm sure 5.5B tokens would have costs more than $6k. Are you paying out of your own pocket or not?

    Beauty of Claude Code Max subscriptions I export my usage metrics Claude Code native OpenTelemetry to Grafana and built my own Grafana MCP so Claude Code can query its own usage over time https://github.com/centminmod/claude-code-opentelemetry-setup

    Example dashboard just after promo for past week https://www.threads.com/@george_sl_liu/post/DWeICtdE3xf?xmt=AQF0JxPCBiDNNjoh9f8IMh__ol2e5sRgFlVJ7AJGpVrwLIH0ZgRp7XMzHRxihY-4fvzSPr5F&slof=1

  • rpqurpqu Member

    @eva2000 said:

    @rpqu said:

    @eva2000 said:
    Claude Code is primary used 5.5 billion tokens the past 30 days of which 3.3 billion tokens during recent 2x off-peak usage 2 week promo.

    The date is April 1 2026, not early 2027. So, I'm sure 5.5B tokens would have costs more than $6k. Are you paying out of your own pocket or not?

    Beauty of Claude Code Max subscriptions I export my usage metrics Claude Code native OpenTelemetry to Grafana and built my own Grafana MCP so Claude Code can query its own usage over time https://github.com/centminmod/claude-code-opentelemetry-setup

    Example dashboard just after promo for past week https://www.threads.com/@george_sl_liu/post/DWeICtdE3xf?xmt=AQF0JxPCBiDNNjoh9f8IMh__ol2e5sRgFlVJ7AJGpVrwLIH0ZgRp7XMzHRxihY-4fvzSPr5F&slof=1

    Ah, you incorporated cache hit into the counter

Sign In or Register to comment.