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.
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.
In addition to usage distribution aspects others called out .
$1K is not actual cost, just API pricing being compared to subscription pricing. It is quite possible that API has a large operating margins, and say costs only $100 to deliver $1K worth of API credits.
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.
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.
I've been thinking of doing this — using one of the "pretty good but not Opus 4.6-good, YET very cheap" models for the implementation part of more basic code features, AFTER first using Opus 4.6 high for the planning stage.
Do you think this would be a decent approach?
Also, which client would I use for this? OpenCode? I don't think Claude Code supports using other models. Thoughts?
I have been doing this and the results have been fairly good.
I use claude to build requirements.md -> implementation.md -> todo.md. Then I tell opencode + openrouter to read those files and follow the todo using a cheap (many times free) model.
It works 90% of the time. The other 10% it will get stuck, in which case I revert to claude.
That has allowed me to stay on the $20/month claude subscription as opposed to the $100.
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.
My Claude Code usage stats after ~3 months of heavy use:
Favorite model: Opus 4.6 Total tokens: 42.6m
Sessions: 420 Longest session: 10d 2h 13m
Active days: 53/95 Longest streak: 16 days
Most active day: Feb 9 Current streak: 4 days
~158x more tokens than Moby-Dick
Monthly breakdown via claude-code-monitor (not sure how accurate this is):
Month Total Tokens Cost (USD)
2026-01 96,166,569 $112.66
2026-02 340,158,917 $393.44
2026-03 2,183,154,148 $3,794.51
2026-04 1,832,917,712 $3,412.72
─────────────────────────────────────
Total 4,452,397,346 $7,713.34
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.
Ignoring the training costs, the marginal cost for inference is pretty low for providers. They are estimated to break even or better with their $20/month subscriptions.
That being said, they can't stop launching new models, so training is not a one time task. Therefore one might argue that it is part of the marginal cost.
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.
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..."
I don't think it's conscious or even a result of not caring about UX/DX. But I do think you're right - I've noticed the loudest voices in their Issue queue are people wanting things like better vim support, helix keybind support (super niche terminal modal editor), etc. Fine if they want to make that their niche but if you are migrating from VS Code like 99% of people you can't have these kinds of papercuts, people will just uninstall.
I think explicit post-training is going to be needed to make this kind of approach effective.
As this repo notes is "The secret to good memory isn't remembering more. It's knowing what to forget." But knowing what is likely to be important in the future implies a working model of the future and your place in it. It's a fully AGI complete problem: "Given my current state and goals, what am I going to find important conditioned on the likelihood of any particular future...". Anyone working with these agents knows they are hopelessly bad at modeling their own capabilities much less projecting that forward.
What is Cursor doing? They need to relax a little bit. Recently I saw they released "Glass" which WAS here: https://cursor.com/glass, now just redirects to /download.
Is "Cursor 3" == Glass? I get they feel like their identity means they need to constantly be pushing the envelope in terms of agent UX. But they could stand to have like an "experimental" track and a "This is VS Code but with better AI integration" track.
Glass was a codename while the UI was in early alpha with testers. It redirects to download now because there is no special link anymore. It's just part of Cursor 3 itself.
If you actually care about this stuff you are going to run something like https://github.com/waydabber/BetterDisplay which easily allows for HiDPI @ 4K resolution, it does not "look bizarre" or "require fractional scaling". This is what the OP is about. I do the same thing, I run native res w/ HiDPI on a 27" 4K screen as my only monitor, works great.
Sure, and that is the real tragedy here. The person I'm replying to is just pointing out that native support for high res sucks, which is true, but the real problem is what limits there are on 3rd party support.
Well, the only people with any ability to acknowledge it have a massive incentive to do so, and I've been around the block enough times to know that startups will use every trick in the book to paint a rosy financial picture, even when it's extremely misleading or occasionally just straight up lies. In the current climate of AI hype my skepticism is even greater.
The CEO hyping his product and the viability of his business during an interview with Stripe does not, at least to me, qualify as “widely reported and acknowledged”
The OpenCode guys have really surprised me in the way they've reacted to Anthropic shutting down the side-loaded auth scheme. Very petty and bitter. It's clearly just a business decision from Anthropic and a rational one at that, usage subsidization to keep people on the first party product surface is practically the oldest business move in the book and is completely valid.
Ever since the shutdown of the side-load they've been pretty vocally anti-anthropic on twitter. Paranoid that anthropic is going to torpedo them via some backdoor now that they own bun, insinuating that anthropic shut down the auth from a position of weakness since OpenCode is a superior product, etc.
The thing is OpenCode IS a great product, I'm not sure it's "superior", but unfortunately the way things are evolving where the model + harness pairing is so important, it does seem like they are in a similar position to Cursor (and do not have the resources to try to pivot into developing their own foundational model).
I wouldn't call OpenCode a "great" product tbh. It's nice that it's FLOSS of course, but the overall quality is a bit underwhelming and it's clearly possible to build much better open agentic harnesses. It would be nice if more people tried to do this.
I think frankly OpenCode is delusional to think that Anthropic is actually "concerned" with them in any way. Anthropic's concerns at this point are on the geopolitical level. I doubt stamping out ToS-violating usage of their subscription services is even on executive radar. OpenAI only allows it because it's a cheap PR win and they take those where they can get them.
Yeah, I recognized the PR author from Twitter (same avatar) and man he really does come across as incredibly juvenile. Shamelessly talking up OpenAI while shitting on Claude models and the motivation is just so transparent.
Valid question. It's because they have a separate product intended for use with general tools: Their API.
Their subscription plans aren't actually "Claude Code Plans". They're subscription plans for their tool suite, which includes claude code. It's offered at a discount because they know the usage of this customer base.
OpenCode used a private API to imitate Claude Code and connect as if it was an Anthropic product, bypassing the need to pay for the API that was for this purpose.
Anthropic has been consistent on this from the start. The subscription plans were never for general use with other tools. They looked the other way for a while but OpenCode was openly flaunting it, so they started doing detection and blocking.
OpenCode and maintainers have gone on the offense on Twitter with some rather juvenile behavior and now they're trying to cheekily allow a plugin system so they can claim they're not supporting it while very obviously putting work into supporting it.
Most of the anger in this thread comes from people who want their monthly subscription to be usable as a cheaper version of the public API, even though it was never sold as that.
This has been explained many times in this thread. Your subscription to Claude models for use in Claude Code is subsidized. That is, it is only meant to be used with that harness.
When you use that API key with OpenCode, you're circumventing that.
The PS5 is subsidized because the make money with the games.
Printers are subsidized because the make money with the ink.
The API use is subsidized because they make money with Claude Code?
I would understand if Claude Code could only be used with Anthropics API but not the other way around. 1 million tokens is 1 million tokens unless Claude Code is burning tokens and others are more efficient in token use.
I'd say that they want Claude Code to become the standard, so that they can milk corporations on enterprise plans. We individual subscribers are nothing, but we'll go to work and be vocal about specifically having Claude.
Because models are quickly moving toward commoditization, whether the big three like it or not. The differentiator now is tooling around those models. By eliminating OpenCode's auth stuff, they prevent leaking customers onto another platform that allows model choice (they will likely lose paying customers to one of the major inference catalogs like OpenRouter once they move from Claude Code to OpenCode).
Why does Netflix care how the movies they stream to you are consumed? Shouldn't your $8/mo allow you to stream any movie to OpenFlix and consume however you like?
You are also not allowed to show these Netflix movies on a big screen in front of your house and charge people. The 8 dollar are for a specific use case, just like the tokens in the subscription.
If you use Claude through an interface that’s not Claude Code, you’ll only stick with it for as long as it proves itself the best. With other interfaces, you can experiment with multiple models and switch from one to another for different tasks or different periods of time.
Those tokens going to other providers are tokens not going to Anthropic, so they want to lock you in with Claude Code. And it clearly works, since a lot of people swear by it.
because he is giving them at 90% discount in their subscription.
they are more than happy if you use the tokens at api pricing, but when subsidized they want you to use their claude code surface.
On what basis are you assuming that Anthropic committed greater copyright theft than Meta, OpenAI, and Google (not to mention many lesser-known options)?
Source: i run pretty much all of these agents (codex, cc, droid, opencode, amp, etc) side-by-side in agentastic.dev and opencode had basically 0 win-rate over other agents.
Anthropic provides subsidized access to Claude models through Claude Code. It is well understood to be 'a loss leader' so that they can incentivize people to use Claude Code.
OpenCode lets people take the Claude-Code-only-API-Key, and lets them use it in a different harness. Anthropic's preferred way for such interaction is getting a different, Claude API key (and not Claude Code SDK API key).
---
A rough analogy might be something like getting subsidized drinks from a cafe, provided you sit there a eat food. What if someone says, go to the cafe and get free drink and come sit over at our cafe and order food. It is a loose analogy, but you get the idea.
If it wasn't the case, the Claude API pricing would be the same, $200 for unlimited use. But it's metered.
We don't know if Claude Code bleeds money for every user that touches it. Probably not. But the different pricing is a strong enough clue that it's an appeal product with subsidized tokens consumption.
API is intended for a different audience - companies with a big pocket who aren't as price sensitive as private users. So the pricing will be different than for a private subscription.
There is huge value in getting people to subscribe to recurring payments. Giving people a discount to do so makes sense and does not mean that the subscription service loses money.
Is this what the legal request demanded or is this just something that OpenCode is doing out of spite? Seems unclear. To me the meat of this change is that they're removing support for `opencode-anthropic-auth` and the prompt text that allows OpenCode to mimic Claude Code behavior. They have been skirting the intent of the original C&D for awhile now with these auth plugins and prompt text.
Using your API key in third-party harnesses has always been allowed. They just don't like using the subsidized subscription plan outside of first-party harnesses. So this seems to be out of spite
reply