I think it is more likely that if you stick with Claude Code, then you are more likely to stick with Opus/Sonnet, whereas if you use a third party CLI you might be more likely to mix and match or switch away entirely. It's in their interest to get you invested in their tooling.
I really like doing this, be it with OpenCode or Copilot or Cline/RooCode/KiloCode: I do have a Cerebras Code subscription (50 USD a month for a lot of tokens but only an okayish model) whereas the rest I use by paying per-token.
Monthly spend ends up being somewhere between 100-150 USD total, obviously depending on what I do and the proportion of simple vs complex tasks.
If Sonnet isn’t great for a given task, I can go for GPT-5 or Gemini 3.
As a matter of principle, I really would like the flexibility though, as while I love Opus now, who knows which model I will prefer next month.
I just use claude code proxy or litellm and set the ANTHROPIC_BASE_URL to my proxy and chose another LLM.
But also, they're a bit schizophrenic about what they want Claude Code to be, given you can stream JSON to/from Claude Code to use it as a headless backend including with your subscriptions.
I can easily churn through $100 in an 8 hour work day with API billing. $200/month seems like an incredibly good deal, even if they apply some throttling.
It's not unreasonable to assume that without the ability to push haiku use aggresively for summarization, the average user in OC vs CC costs more.
So they all want to be product companies. OpenAI is able to keep raising crazy amounts of capital because they're a product company and the API is a sideshow. Anthropic got squeezed because Altman launched ChatGPT first for free and immediately claimed the entire market, meaning Anthropic became an undifferentiated Bing-like also-ran until the moment they launched Claude Code and had something unique. For consumer use Claude still languishes but when it comes to coding and the enormous consumption programmers rack up, OpenAI is the one cloning Claude Code rather than the other way around.
For Claude Code to be worth anything to Anthropic's investors it must be a product and not just an API pricing tier. If it's a product they have so many more options. They can e.g. include ads, charge for corporate SSO integrations, charge extra for more features, add social features... I'm sure they have a thousand ideas, all of which require controlling the user interface and product surface.
That's the entire reason they're willing to engage in their own market dumping by underpricing tokens when consumed via their CLI/web tooling: build up product loyalty that can then be leveraged into further revenue streams beyond paying for tokens. That strategy doesn't work if anyone can just emulate the Claude Code CLI at the wire level. It'd mean Anthropic buys market share for their own competitors.
N.B. this kind of situation is super common in the tech industry. If you've ever looked at Google's properties you'll discover they're all locked behind Javascript challenges that verify you're using a real web browser. The features and pricing of the APIs is usually very different to what consumers can access via their web browser and technical tricks are used to segment that market. That's why SERP scraping is a SaaS (it's far too hard to do directly yourself at scale, has to be outsourced now), and why Google is suing them for bypassing "SearchGuard", which appears to just be BotGuard rebranded. I designed the first version of BotGuard and the reason they use it on every surface now, and not just for antispam, is because businesses require the ability to segment API traffic that might be generated by competitors from end user/human traffic generated by their own products.
If Anthropic want to continue with this strategy they'll need to do the same thing. They'll need to build what is effectively an anti-abuse team similar to the BotGuard team at Google or the VAC team at Valve, people specialized in client integrity techniques and who have experience in detecting emulators over the network.
I think some people get triggered by the inconsistency in pricing or the idea of having a fixed cost for somewhat vague usage limits.
In practice it’s a great deal for anyone using the model at that level.
With "normal use" you're unlikely to hit the limit though, unless you only use the Opus model and do things in parallell.
Isn't the whole thesis behind LLM coding that you can easily clone the CLI using an LLM? Otherwise what are you paying $200/mo for?
Because they are harvesting all the data they can harvest through CLI to train further models. API access in contrast provides much more limited data
I feel like this is a major area of divergence. The "vibes" are bifurcating between "coding agents are great!" and "coding agents are all hype!", with increasing levels of in-group communication.
How should I, an agent-curious user, begin to unravel this mess if $200 is significantly more than pocket change? The pro-agent camp remarks that these frontier models are qualitatively better and using older/cheaper approaches would give a misleading impression, so "buy the discount agent" doesn't even seem like a reasonable starting point.
Sorry, I don't understand this. Either you're saying
A) Everyone paying $200/mo should now pay $800/mo to match this 20% off figure you're theorizing... or B) Maybe you're implying that the $1,000+ costs are too high and they should be lowered, to like, what, $250/mo? (250 - 20% = $200)
Which confuses me, because neither option is feasible or ever gonna happen.
Instead, they're saying that a 200$/month subscription should pay for something like $250 worth of pay-per-token API tokens, and additionally give preferential pricing for using up more tokens than that.
So, if the normal API pricing were 10$ per million tokens, a 200$ per month subscription should offer 25M tokens for free, and allow you to use more tokens at a 9$/1M token rate. This way, if you used 50M tokens in a month, you'd pay 445$ with a subscription, versus paying 500$ with pay-as-you-go. This is still a good discount, but doesn't create perverse incentives, as the current model does.
Anthropic is futher complicated by mission.
I do, it's called vendor lock-in. The product they're trying to sell is not the $200 subscription, it's the entire Claude Code ecosystem.
For the average person, the words "AI" and "ChatGPT" are synonims. OpenAI's competitors have long conceded this loss, and for the most part, they're not even trying to compete, because it's clear to everyone that there is no clear path to monetization in this market - the average joe isn't going to pay for a $100/mo subscription to ask a chatbot to do their homework or write a chocolate cake recipe, so good luck making money there.
The programming market is an entirely different story, though. It's clear that corporations are willing to pay decent money to replace human programmers with a service that does their work in a fraction of the time (and even the programmers themselves are willing to pay independently to do less work, even if it will ultimately make them obsolete), and they don't care enough about quality for that to be an issue. So everyone is currently racing to capture this potentially profitable market, and Claude Code is Anthropic's take on this.
Simply selling the subscription on its own without any lock-in isn't the goal, because it's clearly not profitable, nor is it currently meant to be, it's a loss leader. The actual goal is to get people invested long-term in the Claude Code ecosystem as a whole, so that when the financial reality catches up to the hype and prices have to go up 5x to start making real money, those people feel compelled to keep paying, instead of seeking out cheaper alternatives, or simply giving up on the whole idea. This is why using the subscription as an API for other apps isn't allowed, why Claude Code is closed source, why it doesn't support third party OpenAI-compatible APIs, and why it reads a file called CLAUDE.md instead of something more generic.
I assume they're embarrassed by it. Didn't one of their devs recently say it's 100% vibe coded?
In any case, another workaround would be using ACP that’s supported by Zed. Let’s editing tools access the power of CLI agents directly.
———
> Anthropic should have open sourced their Claude Code CLI a year ago
https://github.com/anthropics/claude-code
It has been open source for a while now. Probably 4-6 months.
> Anthropic shouldn't have an all-you-can-eat plan for $200 when their pay-as-you-go plan would cost more than $1,000+ for comparable usage. Their subscription plans should just sell you API credits at, like, 20% off.
That's a very odd thing to wish for. I love my subscriptions and wouldn't have it any other way.
Another analogy is that it’s a restaurant that offers delivery and they’re insisting you use their own in house delivery service instead of placing a pickup order and asking your friendly neighbor to pick it up for you on their way back from the office.
You don't "get full" and "get hungry again" by switching UIs. You can consume the same amount whether you switch or you don't switch.
Not sure where that goes in the analogies here but maybe something about smaller plates.
This is actually a compelling argument for Claude Code getting the discount but not extending it to other cases. Claude Code, being subsidized by the company, is incentivized to minimize token usage. Third parties that piggyback on the same flat rate subscription, are not. i.e. Claude code wants you to eat less.
Of course, I don’t believe at all that this is why Anthropic has blocked this use case. But it is a reasonable argument.
Think about a web browser that respects cache lifetimes vs one that downloads everything everytime. As an ISP I'd be more likely to offer you unlimited bandwidth if I knew you were using a caching browser.
Likewise Claude code can optimize how it uses tokens and potentially provide the same benefit with less usage.
This is more like an all-you-can-eat restaurant requiring you to eat with their flimsy plastic forks, forbidding you to bring your own utensils.
Why you have this idea? why they should open source it now?
Then it runs the test, as if I could not do this myself, it reads the output, sometimes very long (so more and more tokens are burned) and so on.
If people had to pay for this cleverness and creativity an API price, they would be a bit shocked and give up quickly CC.
Using Aider with Claude Sonnet I am eating much less tokens than CC does.
Who cares? Just have Claude vibe code it in an afternoon...
I don't normally like to come down on the side of the megabigcorp but in this case anthropic aren't being evil. Not yet anyway.
The key question is about why they want to you to use the CLI. If you're not the customer, you're the product.
There's also a monopolistic aspect to this. Having the best model isn't something over can legally exploit to gain advantage in adjacent markets.
It reeks of "Windows isn't done until Lotus won't run," Windows showing spurious error messages for DR-DOS, and Borland C++ losing to the then-inferior Visual C++ due to late support of new Windows features. And Internet Explorer bundling versus Netscape.
Yes, Microsoft badly wanted you to use Office, Visual C++, MS-DOS, and IE, but using Windows to get that was illegal.
Microsoft lost in court, paid a nominal fine, and executives were crying all the way to the bank.
You are the customer, you're paying them directly.
The generic term is predatory pricing and it's regulated to some extent in pretty much every country. https://en.wikipedia.org/wiki/Predatory_pricing#Legal_aspect...
When carried out at the international level it's known as dumping. The WTO has provisions against it. https://en.wikipedia.org/wiki/Dumping_(pricing_policy)#Legal...
In any case, all this depends on how you define the "market", and the entire market for AI-assisted coding is very nascent and fast-moving to make any reliable calls about dominance at this point.
I was asked for examples of behavior that distorts the market being regulated and provided two of them. There are other examples out there as well.
Both scraping and on-demand agent-driven interactions erode that. So you could look at people doing the same to them as a sort of poetic justice, from a purely moral standpoint at least.
"Should have" for what reason? I would be happy if they open sourced Claude Code, but the reality is that Claude Code is what makes Anthropic so relevant in the programming more, much more than the Claude models themselves. Asking them to give it away for free to their competitors seems a bit much.
Here adoption is a combination on the tool and the model.
If people can’t pay the model to use the tool, they might not use the tool even if it’s better.
That’s what anthropic is doing.
It might be faster, but it’s more expensive.
The only way remains to try and lock consumers into your ecosystem.
Aside: it is pretty easy to let our appreciation* of OSS turn into a kind of confirmation bias about its value to other people/orgs.
* I can understand why people promote OSS from various POVs: ethics, security, end user control, pro-innovation, goodwill, and so on. I value all of these things. But I’m wary of reasoning and philosophies that offer merely binary judgments.
Adoption is how one wins. Look at all the crappy solutions out there that are still around.
It's not unheard of for companies that have strong customer mindshare to find themselves intermediated by competitors or other products to the point that they just became part of the infrastructure and eventually lose that mindshare.
I doubt Anthropic wants to become a swappable backend for the actual thing that developers reach for to do their work (the CLI tool).
Don't get me wrong, I think developers should 100% have the choice of tooling they want to use.
But from a business standpoint I think maintaining that direct or first-party connection to the developer is what Anthropic are trying to protect here.
Also it uses the Claude models but afaik it is constantly changing which one is using depending on the perceived difficulty.
Claude Code does the same. You can disable it in Kiro by specifically setting the model to what you want rather than “auto” using /model.
Tbh I’ve found Kiro to be much better than Claude Code. The actual quality of results seems about the same, but I’ve had multiple instances where Claude Code get stuck because of errors making tool calls whereas Kiro just works. Personally I also just prefer the simplicity of Kiro’s UX over CC’s relative “flashy” TUI.
but Claude Code cannot run without Claude models? What do you mean?
also while I was initially on the “they should open source” boat, and I’m happy Codex CLI did, there are a ton of benefits to keeping to closed source. just look at how much spam and dumb community drama OpenAI employees now have to deal with on GitHub. I increasingly think it’s a smart moved to keep it closed source and iterate without as direct community involvement on the codebase for now
They could close the issues and only allow discussions.
There was a project mentioned here recently that did just that.
*Edit
It was Ghostty,
"Why users cannot create Issues directly" - https://news.ycombinator.com/item?id=46460319
Claude Code feels like my early days when pair programming was all the rage.
If you have the time OpenCode comes the closest and lets you work across providers seamless.
I have the "all-you-can-eat" plan _because_ I know what I'm getting and how much it'll cost me.
I don't see anything wrong with this. It's just a big time-limited amount of tokens you can use. Of course it sucks that it's limited to Claude-Code and Claude.ai. But the other providers have very similar subscriptions. Even the original ChatGTP pro subscription gives you a lot more tokens for the $20 it costs compared to the API cost.
I always assumed tokens over the API cost that much, because that's just what people are willing to pay. And what people are willing to pay for small pay-as-you-go tasks vs large-scale agentic coding just doesn't line up.
And then there's the psychological factor: if Claude messed up and wasted a bunch of tokens, I'm going to be super pissed that those specific tokens will have cost me $30. But when it's just a little blip on my usage limit, I don't really mind.
I had the TaskMaster AI tool hooked up to my Anthropic sub, as well as a couple of other things - Kilo Code and and Roo Code iirc?
From discussions at the time (6 months ago) this "use your Anthropic sub" functionality was listed by at least one of the above projects as "thanks to the functionality of the Anthropic SDK you can now use your sub...." implying it was officially sanctioned rather than via a "workaround".
They want you to use their tool so they can collect data.
That data is critical to improve tool call use (both in correctness but also to improve when the agent chooses to use that tool). It's also important for the context rewrites Claude does. They rewrite your prompt and continuously manage the back-and-forth with the model. So does Cortex, just more aggressively with a more powerful context graph.
What part of “OpenCode broke the TOS of something well defined” makes you think it’s all Anthropic’s fault?
FTFY.
GPU compute cost has falled down in the last two years a lot.
It's not like if it houses some top secret AI models inside of it, and it would make way more sense, and probably expand the capabilities of Claude Code itself. Do they lose out to having OpenAI or other competitors basically stealing their approach?
They definitely want their absolutely proprietary software with sudo privilege on your machine. I wonder why they would want that geeez
- effective moneytizeability of a lot of AI products seem questionable
- so AI cost strongly subsidized in all kinds of ways
- which is causing all kind of strange dynamics and is very much incompatible with "free market self regulation" (hence why a company long term running by investor money _and_ under-pricing any competition which isn't subsidized is theoretically not legal (in the US). Not that the US seem to care to actually run a functioning self regulating free market, even going back as far as Amazone. Turns out moving "state subsidized" to "subsidized by rich" somehow makes it no longer problematic / anti-free-market /non-liberal ... /s))
“$17 Per month with annual subscription discount ($200 billed up front). $20 if billed monthly.”
What are you referring to that’s 10x that price? (Conversely, I’m wondering why Pro costs 1/10 the value of whatever you’re referring to?!?)
Once the limit is reached, you can choose to pay-per-token, upgrade your plan, or just wait until it refreshes. The more expensive subscription variants just contain more tokens, that’s all.
We'll both find out if Pro's token limits are sufficient for spending a weekend on a learning project, and what the fallback looks like.
The model is not a moat
They need to own the point of interaction to drive company valuation. Users can more about tool switching costs that the particular model they use.
> oh no, people are actually buying the loss leader
I'm looking forward to the upcoming reckoning when all these AI companies start actually charging users what the services cost.
Most subscribers dont use up all their allocated tokens. Theyre banning these third parties because they consistently do use all their allocated tokens.
Why should subscribers get your specific discount rather than what anthropic has calculated the discount should be?
Hard disagree. Companies can and do subsidize products to gather market share. It's just a loss leader [1]. The big money for them is likely satisfied software engineers pushing their employers to pay for more Anthropic products in an enterprise setting.
Why is Anthropic offering such favorable pricing to subscribers? I dunno. But they really want you to use the Claude Code™ CLI with that subscription, not the open-source OpenCode CLI. They want OpenCode users to pay API prices, which could be 5x or more.
So, of course, OpenCode has implemented a workaround, so that folks paying "only" $200/month can use their preferred OpenCode CLI at Anthropic's all-you-can-eat token buffet.
https://github.com/anomalyco/opencode/issues/7410#issuecomme...
Everything about this is ridiculous, and it's all Anthropic's fault. Anthropic shouldn't have an all-you-can-eat plan for $200 when their pay-as-you-go plan would cost more than $1,000+ for comparable usage. Their subscription plans should just sell you API credits at, like, 20% off.
More importantly, Anthropic should have open sourced their Claude Code CLI a year ago. (They can and should just open source it now.)