People may feel differently about the fee that OpenRouter takes, but I think the service they provide is worth the extra cost.
Having access to dozens of models through a single API key, tracking cost of each request, being able to run the same request on different models and comparing their results next to each other, separating usages through different API keys, adding your own presets, setting your routing rules...
And once you start using an account with multiple users, it's even more useful to have all those features!
Not relying on a subscription and having the right to do exactly what you want with your API key (using it with any tool/harness...) is also a big plus to me.
I agree with you in certain circumstances, but not really for internal user inference. OpenRouter is great if you need to maintain uptime, but for basic usage (chat/coding/self-agents) you can do all of what you mentioned and more with a LiteLLM instance. The number of companies that send a bill is rarely a concern when it comes to “is work getting done”, but I agree with you that minimizing user friction is best.
For general use, I personally don’t see much justification as to why I would want to pay a per-token fee just to not create a few accounts with my trusted providers and add them to an instance for users. It is transparent to users beyond them having a single internal API key (or multiple if you want to track specific app usage) for all the models they have access to, with limits and logging. They wouldn’t even need to know what provider is hosting the model and the underlying provider could be swapped without users knowing.
It is certainly easier to pay a fee per token on a small scale and not have to run an instance, so less technical users could definitely find advantage in just sticking with OpenRouter.
1. The LLM provider doesn't know it's you (unless you have personally identifiable information in your queries). If N people are accessing GPT-5.x using OpenRouter, OpenAI can't distinguish the people. It doesn't know if 1 person made all those requests, or N.
A lot of inference providers for open models only accept prepaid payments, and managing multiple of those accounts is kind of cumbersome. I could limit myself to a smaller set of providers, but then I'm probably overpaying by more than the 5.5% fee
If you're only using flagship model providers then openrouter's value add is a lot more limited
Does OpenRouter perform better than LiteLLM on integration though? I found using Anthropic's models through a LiteLLM-laundered OpenAI-style API to perform noticably worse than using Anthropic's API directly. So I've scrapped considering LiteLLM as an option. It's also just a buggy mess from trying to use their MCP server. The errors it puts out are meaningless, and the UI behaves oddly even in the happy path (error message colored green with Success: prepended).
But if OpenRouter does better (even though it's the same sort of API layer) maybe it's worth it?
I love Openrouter. The ability to define presets, and the ease of access is well worth the few vs. juggling lots of providers separately. I maintain a few subscriptions too - including the most expensive Claude subscription - but Openrouter handles the rest for me.
Love openrouter
I can use cheap models without having to have an api at a bunch of different providers and can use the expensive models when im in a pinch and am maxed out from claude or codex
On the topic of Zed itself as a VSCode replacement - my experience is mixed. I loved it at first, but with time the papercuts add up. The responsiveness difference isn't that big on my system, but Zed's memory usage (with the TS language server in particular) is scandalous. As far as DX goes it's probably at 85% of the level VSCode provides, but in this space QoL features matter a lot. Oh, and it still can't render emojis in buffers on Linux...
I agree. One of the strangest things I found was “in-built” support for a things like TailwindCSS. The LSP starts showing warning in all HTML/TSX files and confuses me to no end. I know, I can turn it off with the setting, but the choice seems so confusing. Tailwind is just one of the thousands of CSS libraries. Why enable it without any mention in the codebase.
I have actually ditched PyCharm for the snappiness of Zed. But the paper cuts are really adding up.
I've been attempting to use Zed as a VSCode replacement but between the lack of bitmap font support (Criminal in an alleged code editor), and the weird UI choices, it's been hard. I want to love it, but what is "performance" if I have to spend more time working around the UI and lack of features from extensions. Strangest issue I've encountered is the colours being completely wrong when using Wayland.. colours are perfect when ran with Xwayland. I'll give it a plus though for native transparency for the background. Much nicer than having to make the entire window transparent, including the foreground text like with VSCode.
The point about papercuts adding up so resonates with me! I loved Zed initially and did find it more responsive than VS Code, loved the Zed Agent autocomplete, etc. However, I eventually and reluctantly went back to VS Code. The papercut that finally did it for me was [this open bug](https://github.com/zed-industries/zed/issues/36516) because of which I was not able to step into a packaged library's code when I was debugging my own code, this was in Python.
I actually find Zed pretty reasonable in terms of memory usage. But yeah, like you say, there are lots of small UX/DX papercuts that are just unfortunate. In some cases I'm not sure it's even Zed's fault, it's just years and years of expecting things to work a certain way because of VS Code and they work differently in Zed.
Eg: Ctrl+P "Open Fol.." in Zed does not surface "Opening a Folder". Zed doesn't call them folders. You have to know that's called "Workspace". And even then, if you type "Open Work..." it doesn't surface! You have to purposefully start with "work..."
my favorite is Ctrl+Shift+A which lets you search through all available UI actions (hence the A). That's just so helpful if you know the IDE can do something but you forgot where in the menu structure it was. And to top things off, you can also use Ctrl+Shift+A to look up the keyboard shortcuts for every possible action
I think there’s a bug? It used to be memory efficient and now I periodically notice it explodes. Quit and restart fixes it
I don’t have any extensions installed and I’m basically leaving it open, idle, as a note scratch space. I do have projects open with many files but not many actual files are open
I have 4-5 typescript projects and one python opened in Zed at any given time (with a bunch of LSPs, ACPs, opened terminals, etc.) and I see around 1.2 - 1.4gb usage.
I opened just one of the typescript projects inside VSCode and I see something like 1gb (combining the helpers usage). I'm not using it actively, so no extra plugins and so on.
That's on mac, so I guess it may vary on other systems.
I think you are kidding if you think you are going to be remotely approximately the quantity/quality of output you get from a $100/max sub with Zed/Openrouter. I easily get $1K+ of usage out of my $100 max sub. And that's with Opus 4.6 on high thinking.
Out of curiosity, how many tokens are people using? I checked my openrouter activity - I used about 550 million tokens in the last month, 320M with Gemini and 240M with Opus. This cost me $600 in the past 30 days. $200 on Gemini, $400 on Opus.
For personal use I've noticed Claude (via the web-based chat UI) making really bizarre mistakes lately like ignoring input or making completely random assumptions. At work Claude Code has turned into an absolute dog. It fails to follow instructions and builds stuff like a lazy junior developer without any architecture, tests, or verification. This is even with max effort, Opus 4.6, multiple agents, early compaction, etc. I don't know what they did but Anthropic's quality lead has basically evaporated for me. I hope they fix it because I've since adapted my project's Claude artifacts for use with Codex and started using it instead - it feels like Claude Code did earlier this year.
I'd like to give the new GLM models a try for personal stuff.
>And people keep claiming the token providers are running inference at a profit.
Not everyone gets $1K of usage, and you don't know how fat the per-token margins are. It's like saying the local buffet place is losing money because you eat $100 worth of takeout for $30.
Yes and when we say things like that we are not talking about plans. Running inference at a profit means api token use is run profitably. It’s a huge unknown what’s happening at the plan level, we know there is subsidy happening but in aggregate impossible to know if it’s profitable or not.
Yeah — I just created an anthropic API key to experiment with pi, and managed to spend $1 in about 30 minutes doing some basic work with Sonnet.
Extrapolating that out, the subscription pricing is HEAVILY subsidized. For similar work in Claude Code, I use a Pro plan for $20/month, and rarely bang up against the limits.
And it scales up - the $200 plan gets you something like 20x what the Pro plan gets you. I've never come close to hitting that limit.
It's obviously capital-subsidized and so I have zero expectation of that lasting, but it's pretty anti-competitive to Cursor and others that rely on API keys.
According to the meter, I used $15k in tokens with my Max plan (along with $5k of Codex tokens) in the last 30 days. That built an entire working and (lightly) optimized language, parser, compiler, runtime toolchain among other things.
Some of the newer models available on OpenRouter are good, but I agree that none of them are a replacement for Opus 4.6 for coding.
If you're trying to minimize cost then having one of the inexpensive models do exploratory work and simple tasks while going back to Opus for the serious thinking and review is a good hybrid model. Having the $20/month Claude plan available is a good idea even if you're primarily using OpenRouter available models.
I think trying to use anything other than the best available SOTA model for important work is not a good tradeoff, though.
Its really fantastic. I can't imagine why you'd go through the effort using Claude Code with other models when pi is a much better harness. There's tons of extensions already available, and its trivial to prompt an LLM to create your any new extension you want. Lacking creativity and want something from another harness?
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
Maybe in a few years there will be obvious patterns with harnesses having built really optimal flows, but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.
> but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.
That’s what really appeals to me. I’ve been fighting Claude Code’s attempts to put everything in memory lately (which is fine for personal preferences), when I prefer the repo to contain all the actual knowledge and learnings. Made me realise how these micro-improvements could ultimately, some day, lead to lock-in.
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
I bought a $30 Z.ai Coding Plan sub to go with it. 7 million tokens has only gone through 2% of my weekly usage using the GLM-5.1 model. I am pretty happy.
I am only doing single project workflows, but with Z.ai I feel like it opens a whole new door to parallel workflows without hitting usage limits.
Yeah, that's pretty much the lines I was thinking along.
Perhaps use codex for planning / reviews, but otherwise go with z.ai / minimax for actual implementation. Thanks!
Pi is a lot simpler than Claude and a lot more transparent in how it operates.
It's designed to be a small simple core with a rich API which you can use for extensions (providing skills, tools, or just modifying/extending the agent's behaviour).
It's likely that you'll eventually need to find extensions for some extended functionality, but for each feature you can pick the one that fits your need exactly (or just use Pi to hack a new extension).
I really love it. The simplicity is key. The first play project I made with it was a public transport map with GTFS data - click on a stop and get the routes and the timetables for the stop and the surrounding ones. I used Qwen3.5-35B on Mac M1 Max with oMLX. It wrote 98% of the code with very little interaction from me. And very useful is the /tree feature to go back in history when the model is on a wrong track or my instructions where not good enough. I usually work in a two path approach: first let the model explore what it needs to fulfill the task and write it into CONTEXT.md (or any other name to your liking). Then restart the session with the CONTEXT.md. That way you are always nicely operating in 5-15k context, i.e. all is very fast. Create an account for pi (or docker) and make sure it can't walk into other directories - it has bash access.
Add the browser-tools to the skills and load them when useful:
https://github.com/badlogic/pi-skills
No need for database MCP, I use postgres and tell it to use psql.
Occasionally I use prettier to remove indentation - the LLM makes a lot less edit errors that way. Just add the indent back before you commit. Or tell pi to do it.
had been using claude max/opus with pi and the results have been incredible. Having pi write an AGENTS.md and dip your feet into creating your own skills specific to the project.
With the anthropic billing change (not being able to use the max credits for pi) I think I have to cancel - as I'm whirring through credits now.
Going to move to the $250/mo OpenAI codex plan for now.
Reluctantly, the dev seems to have a stinky attitude.
He went on an "OSS vacation", which is perfectly reasonable and said he'd be back on a certain date. I had a PR open for a trivial fix, someone asked when it would land. I shared he was still away. After his return I politely asked, "@badlogic hey, what can we do to progress this? Thanks x"
I then got what I would consider an abusive reply, because he confused me with someone else. In the meantime he extended his vacation. Didn't even think his shitty attitude was worthy of an apology, that HE confused me with someone else.
Now he's seemingly marked anything with my name on as a "clanker", despite all my changes being by hand.
I've been around open source enough to have a thick skin, but when i'm doing something "for fun" and someone treats you like that, i'd rather avoid it as far as possible. I certainly could not in good faith use this project for anything work related.
> Honestly, it seems you are grumpy, so it was probably a good idea to extend that vacation. Being rude just creates a more toxic environment for everyone. Maybe extend that break for the rest of the month and come back nicer? Thanks
Honestly, it seems like both of you were feeling a bit "grumpy" at the moment, but sending passive aggressiveness towards the maintainer you are trying to get to merge your code (or not your code, someone else's code?) seems like a very bold strategy regardless.
You seem to have posted your polite question as a reply to the bot comment which talks about PR #1484 and not your PR. I'd say it's pretty obvious why the maintainer thought you were pushing the bot's PR.
As someone else pointed out cooler heads and less passive aggressive responses would've resolved this issue easily.
I'm pretty conservative when it comes to clearing the context, and I also tend to provide the right files to work on (or at least the right starting point).
I had seen prior to using the model that it starts producing much worse results when the context used is larger, so my usage style probably helps getting better results. I work like this with Claude Code anyway, so it wasn't a big change.
My 50c - ollama cloud 20$. GLM5 and kimi are really competitive models, Ollama usage limits insane high, no limits where to use (has normal APIs), privacy and no logging
Interesting. I've always been turned off by how vague the descriptions of Ollama's limits are for their paid tiers. What sort of work have you been doing with it?
yeah? why do you like that over using GLM5 in a VPS that charges by token use? $20 still cheaper and seamless to set up? how are the tokens per second?
I dislike how Zen (and many similar cases, not picking on Zen here) report being not for profit or transparent, while the auto-recharge mechanism guarantees they are sitting on a float of at least $5 per account, and presumably an average of at least $10. That's something like 50 cents of interest income per year per account. It's not nothing and it's hardly egregious fraud, but I feel if they will do this when it's obvious what they're doing, what other corners might they cut
Honesty as a marketing strategy is really undervalued in cases like this
I mostly use Opus via Copilot with opencode, and I'll tell you, in the past few days, I've had long sessions (almost the whole day) without hitting rate limits. That's very different from Claude Code, which used to rate-limit me before even halfway through the day.
I haven't tried $20 claude code recently, but I've used OpenCode Zen primarily so I can play with opensource/chinese models which are very inexpensive. I'd spend $0.50-$1.00 on a single claude opus 4.6 plan mode run, then have a chinese model execute the plan for like $0.10-$0.15 total. I'd keep context short, constantly start new threads, and get laser focused markdown plans and knowledgebase to be token efficient.
If I just let opencode zen run claude opus to plan and execute, I'd spend $20 in like 5 minutes lol
I tried using OpenRouter for the same kind of development I now do with Anthropic's subscription across Sonnet/Gemini/GPT models and it ended up being 2-3x more expensive than the subscription (which I suspect is heavily subsidized).
It's nice that it works for the author, though, and OpenRouter is pretty nice for trying out models or interacting with multiple ones through a unified platform!
I'm running out of Claude session limits in a single planning + implementation session even when using sonnet for the implementation. This isn't even super complex work - it was refactoring a data model, modifying templates/apis/services, etc. It has also gotten notably more 'lazy' like it updated the data model and not the template until I specifically pointed that out.
My backup has been Opencode + Kimi K2. It's definitely not as strong as even Sonnet but it's pretty fast and is serviceable for basic web app work like the above.
I am in a situation where every sub-folder has its own language server settings, lint settings, etc. VSCode (and forks) can handle this by creating a workspace, adding each folder to the workspace, and having a separate .vscode per-folder. I haven't figured out how to do the same with Zed.
The bursty usage pattern is what kills the subscription value. I hit limits during mid-refactor and there's nothing to do but wait. The worst part is knowing the unused hours during the day are just gone.
OpenRouter credit rollover is the real insight — credits that don't expire vs time windows that reset whether you used them or not. I'm surprised Anthropic hasn't offered a token pack option alongside the subscription.
I don't think there's currently better value than Github's $40 plan which gives you access to GPT5 & Claude variants. It's pay per request so not ideal for back-and-forth but great for building complex features on the cheap compared to paying per token.
Because GH is accessing the API behind the scenes, you should face less degradation when using Sonnet/Opus models compared to a Claude subscription.
Keep a ChatGPT $20 subscription alongside for back-and-forth conversations and you'll get great bang for buck.
I'm still paying the 10$ GH copilot but I don't use it because :
- context is aggressively trimmed compared to CC obviously for cost saving reasons, so the performance is worse
- the request pricing model forces me to adjust how I work
Just these alone are not worth saving the 60$/month for me.
I like the VSCode integration and the MCP/LSP usage surprised me sometimes over the dumb grep from CC. Ironically VSCode is becoming my terminal emulator of choice for all the CLI agents - SSH/container access and the automatic port mapping, etc. - it's more convenient than tmux sessions for me. So Copilot would be ideal for me but yeah it's just tweaked for being budget/broad scope tool rather than a tool for professionals that would pay to get work done.
You can use your GH subscription with a different harness. I'm using opencode with it, it turns GH into a pure token provider. The orchestration (compacting, etc.) is left to the harness.
It turns it into a very good value for money, as far as I'm concerned.
GHCP at least is transparent about the pricing: hit enter on a prompt= one request. CC/Codex use some opaque quota scheme, where you never really know if a request will be 1,2,10% of your hourly max, let alone weekly max.
I've never seen much difference with context ostensibly being shorter in GHCP, all of the models (in any provider) lose the thread well before their window is full, and it seems that aggressive autocompaction is a pretty standard way to help with that, and CC/Codex do it frequently.
I just tried Zed with Gemma 4 to see how it does with local models. Impressive speed and quality for the small model with thinking off (E4B). Very slow for the big model with thinking turned on. We'll see if this is better than my current tools (primary is Codex CLI plus qwen3 coder next) but the first impression is good. Especially nice that it configured all of my ollama models automatically.
Our bank (a major retail bank in UK) is refusing doing business with OpenRouter and OpenRouter issued a refund which we did not request. So something is up. There is that.
I might be paranoid but I feel that access to models will become more constraint in the future as the industry gets more regulated.
Bank refused to provide reasons - even after a formal complaint was raised with them.
We are not the only one. I found other people online experiencing the same issue. It is hard to tell how wide-spread this is but it is strange to say the least.
It should be noted about Openrouter that you aren't allowed to expose the access to end users, it has to be for internal usage only, which can be fatal as they have made waves of account banning lately (without warnings).
You are absolutely allowed to expose access to end users, as long as you continue to abide by terms of service. We have hundreds, if not thousands, of apps built on openrouter that in turn have end users of their own. We showcase many of them on our /apps ranking page!
I was actually wondering about this since I've seen like 3 comments talking about the same thing, would it happen to be related to money laundering due to the availability of the crypto payment method?
When I use the tool ccusage it says I use $600 of usage a month for my $100. I don’t know that this is a good value proposition for me if I want to stay with the same model, half the reason I use Claude code, personally.
The new gimped claude code limits means my claude code spend the last month is $131. It cost me $20. I did an additional spend $5 on extra usage which cost me $5.
While VC's are setting fire to money I am going to warm my hands.
I think it is worth noting that “what they charge for api access” != “marginal cost of inference”. So I don’t think getting i.e. $40 of api usage for $20 would be insane. $131 for $20 does probably mean somebody is losing money though.
I ran this just now and for a small web-app I built I used over $50 in a single day. This was using superpowers plugin and almost exclusively coordinating through Opus. Could I get by with 100$ a month without the subscription? Maybe, but I pay for the convenience of just being able to throw Opus with lavish plugins at it (with 5h limits that are, in my opinion, pretty reasonable). I don't really WANT to have to think about when Haiku or Sonnet are enough.
If anything I would consider switching to OpenAI subscription (if I didn't despise them even more than Anthropic as a company), but converting to API use seems completely infeasible to me. I'd have to severely cut back on my use for not much benefit, other than having maybe an agent thats a little less jank than CC.
Depending on your workflow, in the spirit of reallocating $100/Month subscription, it may be worth dropping to the $20/Month plan (or equivalent at other providers) and then pay as you go on the (rare) occasions you "build a small web-app I built and used over $50 in a single day".
But at that point we are just min/maxing the details, and all I can say is if you are on a $100/$200 a month subscription to any of these services and not using them regularly then you shouldn't be on a $200 subscription any more than you should be on a $700 a month gym membership when you go every 3 months for 15 minutes.
I like and do use Zed but be aware functionality like Hooks is not supported for their integration with Claude Code, as a heavy user of Hooks I would stick with the terminal.
I'm always interested in how people use tools. I like to have a full editor to review code as a complement to the CLI and as I don't often use hooks the integration is also good enough for me.
1. What do you use the hooks for?
2. Do you use an editor alongside the CLI to review code or only examine the diffs?
I have had credits on open router that haven’t been deleted since near the projects launch, I believe 365 days is not a rule but rather a right reserved.
COO of OpenRouter here. Thats right — we haven’t done it to date but we can’t have unlimited liabilities stacking up forever. At some point we will start expiring credits from accounts that have seen zero activity in over a year.
Maybe a bad suggestion, but can you do an inactivity "fee" - 25% / year (min $5) or something similar. I like the pre-pay system everyone in Ai seems to have settled on, its better than the AWS bills that we all know and love.
> we can’t have unlimited liabilities stacking up forever
The liabilities are completely offset by prepayments from your customers though. Even better, you can earn interest on the deposits without paying any out.
If you just dont want the liabilities on the books, issue refunds. Expiring credits feels like a cash grab.
Thank you for taking the time to explain that - makes sense. I lifted what was present in your terms of service as I'd like to understand the minimum time I have.
You are absolutely correct, I was not aware of this. I will update the article accordingly and perhaps it's more worthwhile to stay solely on Cursor with the limited models.
Sadly Zed seems to add 10% so it's still more worthwhile to use OpenRouter.
I would suggest to explore paid plans on different providers. Much better value than plans bundled with editors or API based usage in openrouter. And Chinese companies have versions hosted in Singapore or US.
Also ditching Claude Code is mistake. It is quite capable model, and still great value. I would keep it, even if it's just for code reviews and planning. Anthropic allows pro plans use in Zed.
I just so happen to be doing a price comparison for different cloud LLM providers right now. It turns out some of the cheapest providers with the highest limits are ones you might not have heard of.
OpenCode Go has the simplest plan at the highest rate limits for any subscription plan with multiple model families, and it's $10/month ($5/month for first month). With the cheapest model in the plan (MiniMax M2.5), it is a 13x higher rate than Claude Max, at 1/10th the price. The most expensive model (GLM 5.1) gives you a rate of 880 per 5h, which is more than any other $10 plan. I don't expect this price to last, it makes no sense. OpenCode also has a very generous free tier with higher rates than some paid plans, but the free models do collect data.
The cheapest plan of all is free and unlimited - GitHub Copilot. They offer 3 models for free with (supposedly) no limit - GPT-4o, GPT-4.1, and GPT-5-mini. I would not suggest coding with them, but for really basic stuff, you can't get better than free. I would not recommend their paid plans, they actually have the lowest limits of any provider. They also have the most obtuse per-token pricing of any provider. (FYI, GitHub Copilot OAuth is officially supported in OpenCode)
The next cheapest unlimited plan is BlackBox Pro. Their $10/month Pro plan provides unlimited access to MiniMax M2.5. This model is good enough for coding, and the unlimited number of requests means you can keep churning with subagents long after other providers have hit a limit.
The next cheapest is MiniMax Max, a plan from the makers of MiniMax. For $50/month, you get 15,000 requests per 5-hours to MiniMax M2.7. This is not as cheap as OpenCode Go, which gives you 20,000 requests of MiniMax M2.5 for $10, but you are getting the newer model.
If you don't want to use MiniMax, the next cheapest is Chutes Pro. For $20/month, you get a monthly limit of 5,000 requests.
Note: This calculation is inaccurate, for multiple reasons. For one, it's entirely predicated on working 8 hours a day, 22 days a month; I'll recalculate at some point to find cheapest if you wanted to churn 24/7. For another, some providers (coughANTHROPIC) don't actually tell you what their limits are, so we have to guess and use an average. But based on my research, the calculations seems to match up with the per-request API cost reported at OpenRouter. Happy to take suggestions on improvements.
I have been so far happy with the value that Copilot brought but for the past few weeks I have felt the chokehold on the number of requests.
I have had the chance to test the main Chinese models through OpenRouter but the Pay-as-you-go model is expensive compared to a subscription model, but I don't want to marry to a single provider.
Thanks for bringing OpenCode Go to my attention. Your comparison is the research I didn't know I needed, and I will be cancelling my Copilot subscription to replace it with OpenCode Go right away.
Get a Gemini subscription and pipe the antigravity tokens into claude code. You can have five family accounts on one subscription and every account gets the same amount of tokens. It's the best value there is atm and you get more claude tokens than from anthropic themselves.
Sorry can you expand on that? I have a Gemini subscription from a Google pro account but never used it much. I can use it with Claude Code?? Hmm. I’ll look it up. Thanks!
I am very disappointed that Anthropic killed the use of Max subscriptions for OpenClaw, especially when I never hit my usage limits on it. Perhaps I will try this combo as an alternative.
Rather than trying to lie and get people to use your service, be honest what the upsides/downsides are, and only add your spam when it's at least a bit related, otherwise it just comes off as insincere when you're spamming your own platform in unrelated threads.
Hearing my "before I die" math and code ambitions, a psychiatrist friend tried to convince me to hire a full-time programming assistant. Then came AI.
Money is relative. I retired at less than the average professor salary (all ages) at a not-rich school. I would have made more in tech. I still have weeks where the market goes up 2000x my AI budget, just the retirement savings from my salary. Anyone who isn't living in a van and eating peanut butter if they must, to save the max toward retirement, isn't recognizing how profoundly our system is rigged to favor saving.
When you consider the cross section of the tech community posting on HN, is it really that surprising?
It’s mad for sure, but I’d bet 99.9% of people spending money on AI aren’t spending their own hard earned sooo… “YOLO it’s a business expense/investment”…
People may feel differently about the fee that OpenRouter takes, but I think the service they provide is worth the extra cost.
Having access to dozens of models through a single API key, tracking cost of each request, being able to run the same request on different models and comparing their results next to each other, separating usages through different API keys, adding your own presets, setting your routing rules...
And once you start using an account with multiple users, it's even more useful to have all those features!
Not relying on a subscription and having the right to do exactly what you want with your API key (using it with any tool/harness...) is also a big plus to me.
I agree with you in certain circumstances, but not really for internal user inference. OpenRouter is great if you need to maintain uptime, but for basic usage (chat/coding/self-agents) you can do all of what you mentioned and more with a LiteLLM instance. The number of companies that send a bill is rarely a concern when it comes to “is work getting done”, but I agree with you that minimizing user friction is best.
For general use, I personally don’t see much justification as to why I would want to pay a per-token fee just to not create a few accounts with my trusted providers and add them to an instance for users. It is transparent to users beyond them having a single internal API key (or multiple if you want to track specific app usage) for all the models they have access to, with limits and logging. They wouldn’t even need to know what provider is hosting the model and the underlying provider could be swapped without users knowing.
It is certainly easier to pay a fee per token on a small scale and not have to run an instance, so less technical users could definitely find advantage in just sticking with OpenRouter.
The two things I like about OpenRouter:
1. The LLM provider doesn't know it's you (unless you have personally identifiable information in your queries). If N people are accessing GPT-5.x using OpenRouter, OpenAI can't distinguish the people. It doesn't know if 1 person made all those requests, or N.
2. The ability to ensure your traffic is routed only to providers that claim not to log your inputs (not even for security purposes): https://openrouter.ai/docs/guides/routing/provider-selection...
It's been forever since I played with LiteLLM. Can I get these with it?
6 replies →
> The number of companies that send a bill is rarely a concern
Not true in any non startup where there is an actual finance department
LiteLLM had a major security incident recently, and often isn't actually that useful an abstraction...
A lot of inference providers for open models only accept prepaid payments, and managing multiple of those accounts is kind of cumbersome. I could limit myself to a smaller set of providers, but then I'm probably overpaying by more than the 5.5% fee
If you're only using flagship model providers then openrouter's value add is a lot more limited
3 replies →
Does OpenRouter perform better than LiteLLM on integration though? I found using Anthropic's models through a LiteLLM-laundered OpenAI-style API to perform noticably worse than using Anthropic's API directly. So I've scrapped considering LiteLLM as an option. It's also just a buggy mess from trying to use their MCP server. The errors it puts out are meaningless, and the UI behaves oddly even in the happy path (error message colored green with Success: prepended).
But if OpenRouter does better (even though it's the same sort of API layer) maybe it's worth it?
Doesnt Copilot Pro+ does the same?
I love Openrouter. The ability to define presets, and the ease of access is well worth the few vs. juggling lots of providers separately. I maintain a few subscriptions too - including the most expensive Claude subscription - but Openrouter handles the rest for me.
Love openrouter I can use cheap models without having to have an api at a bunch of different providers and can use the expensive models when im in a pinch and am maxed out from claude or codex
well worth the 5% they take
You can get the same with kilo gateway without the fee.
Expect you don't have the right to do what you want with the API Key (see waves of ban lately, many SaaS services have closed because of it).
Unless you provide some more details, at least outline what "do what you want" was in your case, this seems like just straight up FUD.
10 replies →
On the topic of Zed itself as a VSCode replacement - my experience is mixed. I loved it at first, but with time the papercuts add up. The responsiveness difference isn't that big on my system, but Zed's memory usage (with the TS language server in particular) is scandalous. As far as DX goes it's probably at 85% of the level VSCode provides, but in this space QoL features matter a lot. Oh, and it still can't render emojis in buffers on Linux...
I agree. One of the strangest things I found was “in-built” support for a things like TailwindCSS. The LSP starts showing warning in all HTML/TSX files and confuses me to no end. I know, I can turn it off with the setting, but the choice seems so confusing. Tailwind is just one of the thousands of CSS libraries. Why enable it without any mention in the codebase.
I have actually ditched PyCharm for the snappiness of Zed. But the paper cuts are really adding up.
I've been attempting to use Zed as a VSCode replacement but between the lack of bitmap font support (Criminal in an alleged code editor), and the weird UI choices, it's been hard. I want to love it, but what is "performance" if I have to spend more time working around the UI and lack of features from extensions. Strangest issue I've encountered is the colours being completely wrong when using Wayland.. colours are perfect when ran with Xwayland. I'll give it a plus though for native transparency for the background. Much nicer than having to make the entire window transparent, including the foreground text like with VSCode.
The point about papercuts adding up so resonates with me! I loved Zed initially and did find it more responsive than VS Code, loved the Zed Agent autocomplete, etc. However, I eventually and reluctantly went back to VS Code. The papercut that finally did it for me was [this open bug](https://github.com/zed-industries/zed/issues/36516) because of which I was not able to step into a packaged library's code when I was debugging my own code, this was in Python.
I actually find Zed pretty reasonable in terms of memory usage. But yeah, like you say, there are lots of small UX/DX papercuts that are just unfortunate. In some cases I'm not sure it's even Zed's fault, it's just years and years of expecting things to work a certain way because of VS Code and they work differently in Zed.
Eg: Ctrl+P "Open Fol.." in Zed does not surface "Opening a Folder". Zed doesn't call them folders. You have to know that's called "Workspace". And even then, if you type "Open Work..." it doesn't surface! You have to purposefully start with "work..."
The issues you described show a critical lack of awareness from the Zed developers that people migrate to their IDE mainly from VS Code.
They are blowing their "weirdness budget" on nonsense.
QoL features is where WebStorm shines! I don't look forward to when I have to open vscode instead sometimes.
Just the floating and ephemeral "Search in files" modal in Jetbrain IDEs would convince me to switch from any other IDE.
my favorite is Ctrl+Shift+A which lets you search through all available UI actions (hence the A). That's just so helpful if you know the IDE can do something but you forgot where in the menu structure it was. And to top things off, you can also use Ctrl+Shift+A to look up the keyboard shortcuts for every possible action
I still debate how much productivity I've gained from better AI compared to the loss from switching off WebStorm
But their tab complete situation is abysmal, and Supermaven got macrophaged by Cursor
Same here: I found the multibuffers feature really useful, but the extension system really couldn't hold a candle to VS Code at the time of my testing
Spent a couple of hours trying to make the Svelte extension ignore a particular type of false positive CSS error, failed, and returned to VS Code
Will definitely give it another chance when the extension system is more mature though!
I think there’s a bug? It used to be memory efficient and now I periodically notice it explodes. Quit and restart fixes it
I don’t have any extensions installed and I’m basically leaving it open, idle, as a note scratch space. I do have projects open with many files but not many actual files are open
Anyway idk
I have 4-5 typescript projects and one python opened in Zed at any given time (with a bunch of LSPs, ACPs, opened terminals, etc.) and I see around 1.2 - 1.4gb usage.
I opened just one of the typescript projects inside VSCode and I see something like 1gb (combining the helpers usage). I'm not using it actively, so no extra plugins and so on.
That's on mac, so I guess it may vary on other systems.
I think you are kidding if you think you are going to be remotely approximately the quantity/quality of output you get from a $100/max sub with Zed/Openrouter. I easily get $1K+ of usage out of my $100 max sub. And that's with Opus 4.6 on high thinking.
Out of curiosity, how many tokens are people using? I checked my openrouter activity - I used about 550 million tokens in the last month, 320M with Gemini and 240M with Opus. This cost me $600 in the past 30 days. $200 on Gemini, $400 on Opus.
For personal use I've noticed Claude (via the web-based chat UI) making really bizarre mistakes lately like ignoring input or making completely random assumptions. At work Claude Code has turned into an absolute dog. It fails to follow instructions and builds stuff like a lazy junior developer without any architecture, tests, or verification. This is even with max effort, Opus 4.6, multiple agents, early compaction, etc. I don't know what they did but Anthropic's quality lead has basically evaporated for me. I hope they fix it because I've since adapted my project's Claude artifacts for use with Codex and started using it instead - it feels like Claude Code did earlier this year.
I'd like to give the new GLM models a try for personal stuff.
> I easily get $1K+ of usage out of my $100 max sub. And that's with Opus 4.6 on high thinking.
And people keep claiming the token providers are running inference at a profit.
>And people keep claiming the token providers are running inference at a profit.
Not everyone gets $1K of usage, and you don't know how fat the per-token margins are. It's like saying the local buffet place is losing money because you eat $100 worth of takeout for $30.
2 replies →
Yes and when we say things like that we are not talking about plans. Running inference at a profit means api token use is run profitably. It’s a huge unknown what’s happening at the plan level, we know there is subsidy happening but in aggregate impossible to know if it’s profitable or not.
Yeah — I just created an anthropic API key to experiment with pi, and managed to spend $1 in about 30 minutes doing some basic work with Sonnet.
Extrapolating that out, the subscription pricing is HEAVILY subsidized. For similar work in Claude Code, I use a Pro plan for $20/month, and rarely bang up against the limits.
And it scales up - the $200 plan gets you something like 20x what the Pro plan gets you. I've never come close to hitting that limit.
It's obviously capital-subsidized and so I have zero expectation of that lasting, but it's pretty anti-competitive to Cursor and others that rely on API keys.
I ran ccusage on my work Max account and I spend what would cost $300 a week if it was billed at API rates.
According to the meter, I used $15k in tokens with my Max plan (along with $5k of Codex tokens) in the last 30 days. That built an entire working and (lightly) optimized language, parser, compiler, runtime toolchain among other things.
Some of the newer models available on OpenRouter are good, but I agree that none of them are a replacement for Opus 4.6 for coding.
If you're trying to minimize cost then having one of the inexpensive models do exploratory work and simple tasks while going back to Opus for the serious thinking and review is a good hybrid model. Having the $20/month Claude plan available is a good idea even if you're primarily using OpenRouter available models.
I think trying to use anything other than the best available SOTA model for important work is not a good tradeoff, though.
Not everyone is just vibecoding everything and relying on agents running sota models to do anything tho.
Has anyone (other than OpenClaw) used pi? (https://shittycodingagent.ai/, https://pi.dev/)
Any insights / suggestions / best practices?
Its really fantastic. I can't imagine why you'd go through the effort using Claude Code with other models when pi is a much better harness. There's tons of extensions already available, and its trivial to prompt an LLM to create your any new extension you want. Lacking creativity and want something from another harness?
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
Maybe in a few years there will be obvious patterns with harnesses having built really optimal flows, but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.
> but right now it works so much better to experiment and try new approaches and prompts and flows, and pi is the easiest one to tweak and make it your own.
That’s what really appeals to me. I’ve been fighting Claude Code’s attempts to put everything in memory lately (which is fine for personal preferences), when I prefer the repo to contain all the actual knowledge and learnings. Made me realise how these micro-improvements could ultimately, some day, lead to lock-in.
> Run <other harness> in tmux and interrogate it how feature X works, then build me the equivalent as a pi extension.
I’ll give it a try!
Yes, it's super cool. Check Mario's latest talk: https://www.youtube.com/watch?v=Dli5slNaJu0 Armin also has some videos covering it on his channel: https://www.youtube.com/@ArminRonacher/ Pi's Discord is still nice, even though it was a bit flooded after the openclaw thing.
I bought a $30 Z.ai Coding Plan sub to go with it. 7 million tokens has only gone through 2% of my weekly usage using the GLM-5.1 model. I am pretty happy.
I am only doing single project workflows, but with Z.ai I feel like it opens a whole new door to parallel workflows without hitting usage limits.
Yeah, that's pretty much the lines I was thinking along. Perhaps use codex for planning / reviews, but otherwise go with z.ai / minimax for actual implementation. Thanks!
Pi is a lot simpler than Claude and a lot more transparent in how it operates.
It's designed to be a small simple core with a rich API which you can use for extensions (providing skills, tools, or just modifying/extending the agent's behaviour).
It's likely that you'll eventually need to find extensions for some extended functionality, but for each feature you can pick the one that fits your need exactly (or just use Pi to hack a new extension).
I really love it. The simplicity is key. The first play project I made with it was a public transport map with GTFS data - click on a stop and get the routes and the timetables for the stop and the surrounding ones. I used Qwen3.5-35B on Mac M1 Max with oMLX. It wrote 98% of the code with very little interaction from me. And very useful is the /tree feature to go back in history when the model is on a wrong track or my instructions where not good enough. I usually work in a two path approach: first let the model explore what it needs to fulfill the task and write it into CONTEXT.md (or any other name to your liking). Then restart the session with the CONTEXT.md. That way you are always nicely operating in 5-15k context, i.e. all is very fast. Create an account for pi (or docker) and make sure it can't walk into other directories - it has bash access. Add the browser-tools to the skills and load them when useful: https://github.com/badlogic/pi-skills
No need for database MCP, I use postgres and tell it to use psql.
Occasionally I use prettier to remove indentation - the LLM makes a lot less edit errors that way. Just add the indent back before you commit. Or tell pi to do it.
had been using claude max/opus with pi and the results have been incredible. Having pi write an AGENTS.md and dip your feet into creating your own skills specific to the project.
With the anthropic billing change (not being able to use the max credits for pi) I think I have to cancel - as I'm whirring through credits now.
Going to move to the $250/mo OpenAI codex plan for now.
I was looking into this as well since Claude models are costing too much with the Extra Usage changes.
Is OpenAI codex not also charging by usage instead of subscription when using pi?
1 reply →
i really have been enjoying pi a lot
at first i thought i was goring to build lots of extra plugins and commands but what ended up working for me is:
- i have a simpel command that pulls context from a linear issue
- simple review command
- project specific skills for common tasks
I use Daniel Meissler’s PAI and it’s been an incredible harness.
Reluctantly, the dev seems to have a stinky attitude.
He went on an "OSS vacation", which is perfectly reasonable and said he'd be back on a certain date. I had a PR open for a trivial fix, someone asked when it would land. I shared he was still away. After his return I politely asked, "@badlogic hey, what can we do to progress this? Thanks x"
I then got what I would consider an abusive reply, because he confused me with someone else. In the meantime he extended his vacation. Didn't even think his shitty attitude was worthy of an apology, that HE confused me with someone else.
https://github.com/badlogic/pi-mono/discussions/1475#discuss...
And another other thing I fixed with no attribution, just landed it himself separately. https://github.com/badlogic/pi-mono/discussions/1080
and
https://github.com/badlogic/pi-mono/issues/1079#event-223896...
Now he's seemingly marked anything with my name on as a "clanker", despite all my changes being by hand.
I've been around open source enough to have a thick skin, but when i'm doing something "for fun" and someone treats you like that, i'd rather avoid it as far as possible. I certainly could not in good faith use this project for anything work related.
> Honestly, it seems you are grumpy, so it was probably a good idea to extend that vacation. Being rude just creates a more toxic environment for everyone. Maybe extend that break for the rest of the month and come back nicer? Thanks
Honestly, it seems like both of you were feeling a bit "grumpy" at the moment, but sending passive aggressiveness towards the maintainer you are trying to get to merge your code (or not your code, someone else's code?) seems like a very bold strategy regardless.
6 replies →
You seem to have posted your polite question as a reply to the bot comment which talks about PR #1484 and not your PR. I'd say it's pretty obvious why the maintainer thought you were pushing the bot's PR.
As someone else pointed out cooler heads and less passive aggressive responses would've resolved this issue easily.
But hey, the dev was generous to give it an MIT license, you could always just fork it and what you like ¯\_(ツ)_/¯
After hitting Claude limits today I spent the afternoon using OpenCode + GLM 5.1 via OpenRouter and I was very impressed.
OpenCode picked up my CLAUDE.md files and skills straight away, and I got similar performance to Opus 4.6.
How much did it cost for how long?
~$1/hr over 4 hours.
I'm pretty conservative when it comes to clearing the context, and I also tend to provide the right files to work on (or at least the right starting point).
I had seen prior to using the model that it starts producing much worse results when the context used is larger, so my usage style probably helps getting better results. I work like this with Claude Code anyway, so it wasn't a big change.
https://z.ai/subscribe
Many of us got the annual Lite plan when they had the $28 discount. But even at $120 I think it's a good deal.
3 replies →
[dead]
My 50c - ollama cloud 20$. GLM5 and kimi are really competitive models, Ollama usage limits insane high, no limits where to use (has normal APIs), privacy and no logging
Interesting. I've always been turned off by how vague the descriptions of Ollama's limits are for their paid tiers. What sort of work have you been doing with it?
yeah? why do you like that over using GLM5 in a VPS that charges by token use? $20 still cheaper and seamless to set up? how are the tokens per second?
I also dropped Claude Code Max.
I switched to OpenCode Zen + GitHub Copilot. For some reason, Claude Code burns through my quota really quickly.
https://opencode.ai/zen
I dislike how Zen (and many similar cases, not picking on Zen here) report being not for profit or transparent, while the auto-recharge mechanism guarantees they are sitting on a float of at least $5 per account, and presumably an average of at least $10. That's something like 50 cents of interest income per year per account. It's not nothing and it's hardly egregious fraud, but I feel if they will do this when it's obvious what they're doing, what other corners might they cut
Honesty as a marketing strategy is really undervalued in cases like this
Yeah man, it's a grand scheme to skim 50 cents off you per year. All combined, that might be just enough to cover their website hosting costs.
honestly the issue with Zen is that they collect and might sell your data
How does Claude Code compare to OpenCode Zen? I’m on the $20/month Claude plan, and was considering OpenCode Zen as well.
Due to the quota changes, I actually find myself using Claude less and less
I mostly use Opus via Copilot with opencode, and I'll tell you, in the past few days, I've had long sessions (almost the whole day) without hitting rate limits. That's very different from Claude Code, which used to rate-limit me before even halfway through the day.
1 reply →
I haven't tried $20 claude code recently, but I've used OpenCode Zen primarily so I can play with opensource/chinese models which are very inexpensive. I'd spend $0.50-$1.00 on a single claude opus 4.6 plan mode run, then have a chinese model execute the plan for like $0.10-$0.15 total. I'd keep context short, constantly start new threads, and get laser focused markdown plans and knowledgebase to be token efficient.
If I just let opencode zen run claude opus to plan and execute, I'd spend $20 in like 5 minutes lol
3 replies →
I tried using OpenRouter for the same kind of development I now do with Anthropic's subscription across Sonnet/Gemini/GPT models and it ended up being 2-3x more expensive than the subscription (which I suspect is heavily subsidized).
It's nice that it works for the author, though, and OpenRouter is pretty nice for trying out models or interacting with multiple ones through a unified platform!
I'm running out of Claude session limits in a single planning + implementation session even when using sonnet for the implementation. This isn't even super complex work - it was refactoring a data model, modifying templates/apis/services, etc. It has also gotten notably more 'lazy' like it updated the data model and not the template until I specifically pointed that out.
My backup has been Opencode + Kimi K2. It's definitely not as strong as even Sonnet but it's pretty fast and is serviceable for basic web app work like the above.
Does anyone use Zed with a monorepo?
I am in a situation where every sub-folder has its own language server settings, lint settings, etc. VSCode (and forks) can handle this by creating a workspace, adding each folder to the workspace, and having a separate .vscode per-folder. I haven't figured out how to do the same with Zed.
I would love to stop using VSCode forks
The bursty usage pattern is what kills the subscription value. I hit limits during mid-refactor and there's nothing to do but wait. The worst part is knowing the unused hours during the day are just gone.
OpenRouter credit rollover is the real insight — credits that don't expire vs time windows that reset whether you used them or not. I'm surprised Anthropic hasn't offered a token pack option alongside the subscription.
I really don't like OpenCode. One thing that really irritated me is that on mouse hover it selects options when you're given a set of choices.
I don't think there's currently better value than Github's $40 plan which gives you access to GPT5 & Claude variants. It's pay per request so not ideal for back-and-forth but great for building complex features on the cheap compared to paying per token.
Because GH is accessing the API behind the scenes, you should face less degradation when using Sonnet/Opus models compared to a Claude subscription.
Keep a ChatGPT $20 subscription alongside for back-and-forth conversations and you'll get great bang for buck.
I'm still paying the 10$ GH copilot but I don't use it because :
Just these alone are not worth saving the 60$/month for me.
I like the VSCode integration and the MCP/LSP usage surprised me sometimes over the dumb grep from CC. Ironically VSCode is becoming my terminal emulator of choice for all the CLI agents - SSH/container access and the automatic port mapping, etc. - it's more convenient than tmux sessions for me. So Copilot would be ideal for me but yeah it's just tweaked for being budget/broad scope tool rather than a tool for professionals that would pay to get work done.
You can use your GH subscription with a different harness. I'm using opencode with it, it turns GH into a pure token provider. The orchestration (compacting, etc.) is left to the harness.
It turns it into a very good value for money, as far as I'm concerned.
3 replies →
Disagree entirely.
GHCP at least is transparent about the pricing: hit enter on a prompt= one request. CC/Codex use some opaque quota scheme, where you never really know if a request will be 1,2,10% of your hourly max, let alone weekly max.
I've never seen much difference with context ostensibly being shorter in GHCP, all of the models (in any provider) lose the thread well before their window is full, and it seems that aggressive autocompaction is a pretty standard way to help with that, and CC/Codex do it frequently.
1 reply →
Is your source code worth only $40 for them to train their models on?
https://www.techradar.com/pro/bad-news-skeptics-github-says-...
Considering how much data they already have from everything that's on GitHub, I doubt you would make a dent boycotting their AI product.
2 replies →
Google $20/mo plan has great usage for Claude Opus. Last time I used it, around Feb, it felt basically unlimited.
Agree, that was Feb. Not now, I cancelled mine on the 7th. Claude Opus via Gemini is just a few prompts then it locks you out for another week.
So, you basically tried it a century ago...
I just tried Zed with Gemma 4 to see how it does with local models. Impressive speed and quality for the small model with thinking off (E4B). Very slow for the big model with thinking turned on. We'll see if this is better than my current tools (primary is Codex CLI plus qwen3 coder next) but the first impression is good. Especially nice that it configured all of my ollama models automatically.
Our bank (a major retail bank in UK) is refusing doing business with OpenRouter and OpenRouter issued a refund which we did not request. So something is up. There is that.
I might be paranoid but I feel that access to models will become more constraint in the future as the industry gets more regulated.
I don't quite understand what you mean by something is up. Was the reason around security/telemetry or similar?
Bank refused to provide reasons - even after a formal complaint was raised with them.
We are not the only one. I found other people online experiencing the same issue. It is hard to tell how wide-spread this is but it is strange to say the least.
OpenRouter accepts crypto for payments. That should have raised some flags with banks.
Im using z.ai when I hit my Claude limit after a few questions..drops in easily in Claude code.
>>For some reason Zed limits the Gemini 3.1 context to 200k tokens
It’s not just Zed, CoPilot also reduces the capabilities and options available when using models directly.
No thanks, definitely agree with the Open Router approach or native harness to keep full functionality.
What providers offer nowadays coding plans, so no pricing per tokens, just api call limit and a monthly fee. Which are affordable?
It should be noted about Openrouter that you aren't allowed to expose the access to end users, it has to be for internal usage only, which can be fatal as they have made waves of account banning lately (without warnings).
You are absolutely allowed to expose access to end users, as long as you continue to abide by terms of service. We have hundreds, if not thousands, of apps built on openrouter that in turn have end users of their own. We showcase many of them on our /apps ranking page!
TOS says: access the Site or Service for purposes of reselling API access to AI Models or otherwise developing a competing service;
So yes obviously you can do what you want as long as you abide by terms of service, but the terms of service does NOT allow you to resell the API.
2 replies →
I was actually wondering about this since I've seen like 3 comments talking about the same thing, would it happen to be related to money laundering due to the availability of the crypto payment method?
6 replies →
When I use the tool ccusage it says I use $600 of usage a month for my $100. I don’t know that this is a good value proposition for me if I want to stay with the same model, half the reason I use Claude code, personally.
> Reallocating $100/Month Claude Code Spend
The new gimped claude code limits means my claude code spend the last month is $131. It cost me $20. I did an additional spend $5 on extra usage which cost me $5.
While VC's are setting fire to money I am going to warm my hands.
You mean you were getting more than $130 per $20 before?
85% discount is actually a bit lower than I remember. I think it used to be closer to 90-95%. They're getting stingy ;)
1 reply →
I think it is worth noting that “what they charge for api access” != “marginal cost of inference”. So I don’t think getting i.e. $40 of api usage for $20 would be insane. $131 for $20 does probably mean somebody is losing money though.
I ran this just now and for a small web-app I built I used over $50 in a single day. This was using superpowers plugin and almost exclusively coordinating through Opus. Could I get by with 100$ a month without the subscription? Maybe, but I pay for the convenience of just being able to throw Opus with lavish plugins at it (with 5h limits that are, in my opinion, pretty reasonable). I don't really WANT to have to think about when Haiku or Sonnet are enough.
If anything I would consider switching to OpenAI subscription (if I didn't despise them even more than Anthropic as a company), but converting to API use seems completely infeasible to me. I'd have to severely cut back on my use for not much benefit, other than having maybe an agent thats a little less jank than CC.
Depending on your workflow, in the spirit of reallocating $100/Month subscription, it may be worth dropping to the $20/Month plan (or equivalent at other providers) and then pay as you go on the (rare) occasions you "build a small web-app I built and used over $50 in a single day".
But at that point we are just min/maxing the details, and all I can say is if you are on a $100/$200 a month subscription to any of these services and not using them regularly then you shouldn't be on a $200 subscription any more than you should be on a $700 a month gym membership when you go every 3 months for 15 minutes.
1 reply →
I like and do use Zed but be aware functionality like Hooks is not supported for their integration with Claude Code, as a heavy user of Hooks I would stick with the terminal.
I'm always interested in how people use tools. I like to have a full editor to review code as a complement to the CLI and as I don't often use hooks the integration is also good enough for me.
1. What do you use the hooks for?
2. Do you use an editor alongside the CLI to review code or only examine the diffs?
[flagged]
I have had credits on open router that haven’t been deleted since near the projects launch, I believe 365 days is not a rule but rather a right reserved.
COO of OpenRouter here. Thats right — we haven’t done it to date but we can’t have unlimited liabilities stacking up forever. At some point we will start expiring credits from accounts that have seen zero activity in over a year.
Maybe a bad suggestion, but can you do an inactivity "fee" - 25% / year (min $5) or something similar. I like the pre-pay system everyone in Ai seems to have settled on, its better than the AWS bills that we all know and love.
> we can’t have unlimited liabilities stacking up forever
The liabilities are completely offset by prepayments from your customers though. Even better, you can earn interest on the deposits without paying any out.
If you just dont want the liabilities on the books, issue refunds. Expiring credits feels like a cash grab.
1 reply →
What if I deposited $10, and have lots of recent activity on free models and have barely touched the $10 for payg models?
Thank you for taking the time to explain that - makes sense. I lifted what was present in your terms of service as I'd like to understand the minimum time I have.
In CA gift cards don’t expire and the industry does fine without having people buy expiring money.
[dead]
note: doesn't openrouter charge 5.5% fee?
You are absolutely correct, I was not aware of this. I will update the article accordingly and perhaps it's more worthwhile to stay solely on Cursor with the limited models.
Sadly Zed seems to add 10% so it's still more worthwhile to use OpenRouter.
Look again, they don't charge that fee until after "1M requests per month" whatever that means? Oh that's if you bring your own provider keys.
https://openrouter.ai/docs/guides/overview/auth/byok
I feel like a bit of an idiot because I didn’t know this either. I just assumed OR was another startup burning money to provide models at cost.
OpenRouter is a valuable service but I’ll probably try to run my own router going forward.
Come on at least write the Hackernews replies yourself.
3 replies →
I would suggest to explore paid plans on different providers. Much better value than plans bundled with editors or API based usage in openrouter. And Chinese companies have versions hosted in Singapore or US.
Also ditching Claude Code is mistake. It is quite capable model, and still great value. I would keep it, even if it's just for code reviews and planning. Anthropic allows pro plans use in Zed.
Just on Zed: it's speed and responsiveness are very impressive. Feels as snappy as Notepad++.
I heard you liked men in the middle, so we put a man in the middle of men in the middle.
I just so happen to be doing a price comparison for different cloud LLM providers right now. It turns out some of the cheapest providers with the highest limits are ones you might not have heard of.
OpenCode Go has the simplest plan at the highest rate limits for any subscription plan with multiple model families, and it's $10/month ($5/month for first month). With the cheapest model in the plan (MiniMax M2.5), it is a 13x higher rate than Claude Max, at 1/10th the price. The most expensive model (GLM 5.1) gives you a rate of 880 per 5h, which is more than any other $10 plan. I don't expect this price to last, it makes no sense. OpenCode also has a very generous free tier with higher rates than some paid plans, but the free models do collect data.
The cheapest plan of all is free and unlimited - GitHub Copilot. They offer 3 models for free with (supposedly) no limit - GPT-4o, GPT-4.1, and GPT-5-mini. I would not suggest coding with them, but for really basic stuff, you can't get better than free. I would not recommend their paid plans, they actually have the lowest limits of any provider. They also have the most obtuse per-token pricing of any provider. (FYI, GitHub Copilot OAuth is officially supported in OpenCode)
The next cheapest unlimited plan is BlackBox Pro. Their $10/month Pro plan provides unlimited access to MiniMax M2.5. This model is good enough for coding, and the unlimited number of requests means you can keep churning with subagents long after other providers have hit a limit.
The next cheapest is MiniMax Max, a plan from the makers of MiniMax. For $50/month, you get 15,000 requests per 5-hours to MiniMax M2.7. This is not as cheap as OpenCode Go, which gives you 20,000 requests of MiniMax M2.5 for $10, but you are getting the newer model.
If you don't want to use MiniMax, the next cheapest is Chutes Pro. For $20/month, you get a monthly limit of 5,000 requests.
I'll be adding more of these as I find them to this spreadsheet: https://codeberg.org/mutablecc/calculate-ai-cost/src/branch/...
Note: This calculation is inaccurate, for multiple reasons. For one, it's entirely predicated on working 8 hours a day, 22 days a month; I'll recalculate at some point to find cheapest if you wanted to churn 24/7. For another, some providers (coughANTHROPIC) don't actually tell you what their limits are, so we have to guess and use an average. But based on my research, the calculations seems to match up with the per-request API cost reported at OpenRouter. Happy to take suggestions on improvements.
I have been so far happy with the value that Copilot brought but for the past few weeks I have felt the chokehold on the number of requests.
I have had the chance to test the main Chinese models through OpenRouter but the Pay-as-you-go model is expensive compared to a subscription model, but I don't want to marry to a single provider.
Thanks for bringing OpenCode Go to my attention. Your comparison is the research I didn't know I needed, and I will be cancelling my Copilot subscription to replace it with OpenCode Go right away.
Yep, I was comparing opencode go ($10/month) with copilot pro ($10/month) this morning.
opencode go gives about 14x the requests of copilot pro. I was like, there must be something not right.
Then I compared the best model GLM5.1 on opencode go, and antropic opus 4.6, yes opus is better on most benchmarks, but glm 5.1 is not too far behind.
Get a Gemini subscription and pipe the antigravity tokens into claude code. You can have five family accounts on one subscription and every account gets the same amount of tokens. It's the best value there is atm and you get more claude tokens than from anthropic themselves.
Sounds like a good way to get your google account banned
Sorry can you expand on that? I have a Gemini subscription from a Google pro account but never used it much. I can use it with Claude Code?? Hmm. I’ll look it up. Thanks!
Be aware it's against the terms of service. Google account ban is possible
I am very disappointed that Anthropic killed the use of Max subscriptions for OpenClaw, especially when I never hit my usage limits on it. Perhaps I will try this combo as an alternative.
[dead]
[dead]
[dead]
[dead]
[dead]
[dead]
[flagged]
Also, when you janandonly pay PPQ.AI rather than OpenRouter it seems to go into your pocket instead, so understandable it makes sense it gets cheaper for you. https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
Rather than trying to lie and get people to use your service, be honest what the upsides/downsides are, and only add your spam when it's at least a bit related, otherwise it just comes off as insincere when you're spamming your own platform in unrelated threads.
I can't believe people are spending $100 a month on this! You're all mad!
Hearing my "before I die" math and code ambitions, a psychiatrist friend tried to convince me to hire a full-time programming assistant. Then came AI.
Money is relative. I retired at less than the average professor salary (all ages) at a not-rich school. I would have made more in tech. I still have weeks where the market goes up 2000x my AI budget, just the retirement savings from my salary. Anyone who isn't living in a van and eating peanut butter if they must, to save the max toward retirement, isn't recognizing how profoundly our system is rigged to favor saving.
When you consider the cross section of the tech community posting on HN, is it really that surprising?
It’s mad for sure, but I’d bet 99.9% of people spending money on AI aren’t spending their own hard earned sooo… “YOLO it’s a business expense/investment”…
I had a similar opinion a couple of years ago, content with more of an autocomplete.
Now I'm happy with agents as the models and harnesses have improved significantly but the token usage comes at a cost.
$200 for Claude, $50 to Open AI, and maybe $100 for Openrouter, and a second Claude account paid by a client... Likely to increase.
It easily pays for itself 10x over.
some are spending 100/day or even 1000/day. they must really be mad :)
Drunk on perceived power
1 reply →
To get the equivalent of a junior developer that would cost $80,000/yr + benefits?
[dead]