Why is that their “huge asset?” The genus of this complaint is that Opencode et al replace everything but the LLM, so it seems like the latter is the true “huge asset.”
If Clause Code is being offered at or near operational breakeven, I don’t see the advantage of lock-in. If it’s being offered at a subsidy, then it’s a hint that Claude Code itself is medium-term unsustainable.
“Training data” is a partial but not full explanation of the gap, since it’s not obviously clear to me how Anthropic can learn from Claude Code sessions but not OpenCode sessions.
Right now, most enterprises are experimenting with different LLMs and once they chose they will be locked for a long time. If they cant can't chose because their coding agent doesn't let them they be locked to that.
Making this mistake could end up being the AI equivalent of choosing Oracle over Postgres
It's a supposedly professional tool with a value proposition that requires being in your work flow. Are you going to keep using a power drill on your construction site that bricks itself the last week or two of every month?
An error message says contact support. They then point you to an enterprise plan for 150 seats when you have only a couple dozen devs. Note that 5000 / 25 = 200 ... coincidence? Yeah, you are forbidden to give them more than Max-like $200/dev/month for the usage-based API that's "so expensive".
They are literally "please don't give us money any more this month, thanks".
I imagine a combination of stop loss and market share. If larger shops use up compute, you can't capture as many customers by headcount.
// There was a figure around o3, an astonishing model punching far above the weights (ahem) of models that came after, that suggested the thinkiest mode cost on the order of $3500 to do a deep research. Perhaps OpenAI can afford that, while Anthropic can't.
But this is not the equivalent of Oracle over Postgres, as these are different technology stacks that implement an independent relational database. Here were talking about Opencode which depends on Claude models to work "as a better Claude" (according to the enraged users in the webs). Of course, one can still use OC with a bazillion other models, but Anthropic is saying that if you want the Claude Code experience, you gotta use the CC agent period.
Now put yourself in the Anthropic support person shoes, and suppose you have to answer an issue of a Claude Max user who is mad that OC is throwing errors when calling a tool during a vibe session, probably because the multi-million dollar Sonnet model is telling OC to do something it can't because its not the claude agent. Claude models are fine-tuned for their agent! If the support person replies "OC is an unsupported agent for Claude Code Max" you get an enraged customer anyway, so you might as well cut the crap all together by the root.
The client is closed source for a reason and they issued DMCA takedowns against people who published sourcemaps for a reason.
I wonder when they will add another level and talk to LLM how to talk to another LLM how to talk to another LLM
The CLI tool is terrible compared to opencode.
That is the unfortunate reality, we are now being foisted claude code. :( I wish they just fork opencode.
It's all easily swappable without OpenCode. Just symlink CLAUDE.md -> AGENTS.md and run `codex` instead of `claude`.
> they are working hard on making it impossible for other models to support their every increasing agent feature set (skills, teleport and remote sessions, LSP, Chrome integration, etc).
Every feature you listed has an open-source MCP server implementation, which means every agent that supports MCP already has all those features. MCP is so epic because it has already nailed the commodification coffin firmly shut. Besides, Anthropic has way less funding than OAI or Google. They wouldn't win the moat-building race even if there were one.
That said, the conventional wisdom is that lowering switching costs benefits the underdogs, because the incumbents have more market share to lose.
This is simple and beautiful. Thank you for sharing it :)
is there a way to do this on windows?
Maybe they can hope to murder opencode in the meantime with predatory pricing and build an advantage that they don't currently have. It seems unlikely though - the fact that they're currently behind proves the barrier to building this sort of tool isn't that high, and there's lots of developers who build their own tooling for fun that you can't really starve out of doing that.
I'm not convinced that attempting to murder opencode is a mistake - if you're losing you might as well try desperate tactics. I think the attempt is a pretty clear signal that Antrhopic is losing though.
I mean I guess they could do a bait and switch (drop prices so low that Anthropic goes bankrupt, then raises price) but that’s possible in literally any industry, and sees unlikely given the current number of competitors
They're after the enterprise market - where office / workspace + app + directory integration, security, safety, compliance etc. are more important. 80% of their revenue is from enterprise - less churn, much higher revenue per W/token, better margins, better $/user.
Microsoft adopting the Anthropic models into copilot and Azure - despite being a large and early OpenAI investor - is a much bigger win than yet another image model used to make memes for users who balk at spending $20 per month.
Same with the office connector - which is only available to enterprises[0] (further speaking to where their focus is). There hasn't yet been a "claude code" moment for office productivity, but Anthropic are the closest to it.
[0] This may be a mistake as Claude Code has been adopted from the ground up
Usually you can see it when someone nags about “call us” pricing that is targeted at enterprise. People that nag about it are most likely not the customers someone wants to cater to.
Most of the time you want to cut off ‘non customers’ as soon as possible and don’t leave ‘big fish’ without having direct contact person who can explain stuff. People just clicking around on their own will make assumptions that need to be addressed in a way no one wastes time.
If you mean this literally, one of the best ways to turn non-customers into customers is to give them a way to pay you. Which means telling them the price. If you're implying something else by ‘non customers’ then I'm missing the implication.
> and don’t leave ‘big fish’ without having direct contact person who can explain stuff
You can give a contact person and have a list of prices.
> People just clicking around on their own will make assumptions that need to be addressed in a way no one wastes time.
Making everyone call you to negotiate is going to waste time.
... an extremely popular marketing tool ... sending an equally excessive amount of data above what they were paying for. They were far less adamant about the product, and on some days I didn't even want them as a customer. If there was a minor blip in the service, they were the first to complain. Reminder, [Sentry] was still a side project at the time so I had a day-job. That meant it was often stressful for Chris and I to deal w/ customer support, and way more stressful dealing with outages.
We had one customer who loved the product, and one who didn't. Both of these customers had such extreme volumes of data that it had a tangible infrastructure cost associated with hosting them. We knew the best thing to do was to find a way to be able to charge them more money for the amount of data they sent. So we set off to build that and then followed up with each customer.
To our surprise, the customer that loved the product didn't want to pay more. The customer who was constantly complaining immediately jumped on the opportunity. What's the lesson to take away from this?
... when I was a teenager I worked at Burger King, and there was an anecdote I will never forget: for every customer that complains, there are nine more with a similar experience. I've cemented this in my philosophy around development, to the point where I now believe over rotating on negative feedback is actually just biasing towards the customers who truly see the value in what you're offering. The customer that was complaining really valued our product, whereas the customer that was happy was simply content.
A $7 Subscription, https://cra.mr/a-seven-dollar-subscription / https://archive.vn/IWS0A (2023).I am curious how big of a chance they have. I could imagine many enterprises that are already (almost by default) Microsoft customers (Windows, Office, Entra etc.) will just default to Copilot (and maybe Azure) to keep everything neatly integrated.
So an enterprise would need to be very dedicated to use everything Microsoft, but then go through the trouble use Claude as their AI just because it is slightly better for coding.
I have a feeling I am missing something here though, I would be happy for anyone to educate me!
Can't light a candle to Opus 4.5 who can now create and modify financial models from PDFs and augmented with websearch and the Excel skill (gpt-5.2 can do this too). That said the market IS smaller
Anthropic is rather obnoxious about training on user data, and I wonder if enterprises (and small businesses!) will grow up soon and start using competing products instead.
(Not that Google is amazing in this regard — their purchasable product options are all over the place to the point where it might be nearly a full time (human!) job to keep track of how to correctly purchase Gemini. Gemini itself seems incapable of figuring this out, or at least I haven’t found the right prompt yet. Gemini is absolutely amazing at hallucinating Google product offerings. OpenAI, on the other hand, seems to have nailed this.)
But that means they lose on inference. Which isn't good.
The types of people who would use this tool are precisely the types of people who don't pay for licenses or tools. They're in a race to the bottom and they don't even know it.
> and that's a very thin layer
I don't think Anthropic understands the market they just made massive investments in.
the reason i got the subscription wasnt to use claude code. when i subscribed you couldnt even use it for claude code. i got it because i figured i could use those tokens for anything, and as i figured out useful stuff, i could split it off onto api calls.
now that exploration of "what can i do with claude" will need to be elsewhere, and the results of a working thing will want to stay with the model that its working on.
If Anthropic ended up in a position that they had to beg various Client providers to be integrated (properly) and had to compete with other LLMs on the same clients and could be swapped out at a moment's notice, they would just become a commodity and lose all leverage. They don't want to end up in such situation. They do need to control the delivery of the product end-to-end to ensure that they control the customer relationship and the quality.
This is also going to be KEY in terms of democratizing the AI industry for small startups because this model of ai-outside-tools-inside provides an alternative to tools-outside-ai-inside platforms like Lovable, Base44 and Replit which don't leave as much flexibility in terms of swapping out tooling.
I use CC as my harness but switch between third party models thanks to ccs. If Anthropic decided to stop me from using third party models in CC, I wouldn't just go "oh well, let's buy another $200/mo Claude subscription now". No. I'd be like: "Ok, I invested in CC—hooks/skills/whatever—but now let's ask CC to port them all to OpenCode and continue my work there".
I don't understand, why would other models not be able to support any, or some, or even a particular single one of these? I don't even see most of these as relevant to the model itself, but rather the harness/agentic framework around it. You could argue these require a base degree of model competence for following instructions, tool calling, etc, but these things are assumed for any SOTA model today, we are well past this. Almost all of these things, if not all, are already available in other CLI + IDE-based agentic coding tools.
Claude code runs into use limitations for everyone at every tier. The API is too expensive to use and it's _still_ subsidized.
I keep repeating myself but no one seems to listen: quadratic attention means LLMs will always cost astronomically more than you expect after running the pilot project.
Going from 10k loc to 100k loc isn't a 10x increase, it's a 99x increase. Going from 10k loc to 1m loc isn't a 100x increase, it's a 9999x increase. This is fundamental to how transformers work and is the _best case scenario_. In practice things are worse.
So what you say is not true: cost does not directly correlate with LoC.
What do you mean by this? I know plenty of people who never hit the upgraded Opus 4.5 limits anymore even on the $100 plan, even those who used to hit the limits on the $200 plan w/ Opus 4 and Opus 4.1.
>The API is too expensive to use and it's _still_ subsidized.
What do you mean by saying the API is subsidized? Anthropic is a private company that isn't required to (and doesn't) report detailed public financial statements. The company operating at a loss doesn't mean all inference is operating at a loss, it means that the company is spending an enormous amount of money on R&D. The fact that the net loss is shrinking over time tells us that the inference is producing net profit over time. In this business, there is enormous up front cost to train a model. That model then goes on to generate initially large, but subsequently gradually diminishing revenue until the model is deprecated. That said, at any given snapshot-in-time, while there is likely large ongoing R&D expenditure on the next model causing the overall net profit for the entire company to still be negative, it's entirely possible that several, if not many or even most of the previously trained models have fully recouped their training costs in inference revenue.
It's fairly obvious that the monthly subscriptions are subsidized to gain market share the same way Uber rides were on early on, but what indication do you have that the PAYG API is being subsidized? How would total losses have shrunk from $5.6B in 2024 to just $3B in 2025 while ARR grew from ~$1B to ~$7B over the same time period (one where usage of the platform dramatically expanded) if PAYG API inference wasn't running at a net profit for the company?
>quadratic attention means LLMs will always cost astronomically more than you expect after running the pilot project
This is only true as long as O(n²) quadratic attention remains the prevailing paradigm. As Qwen3-Next and Nemotron 3 Nano have shown with hybrid linear attention + sparse quadratic layers and a hybrid Mamba SSM, not all modern, performant LLMs necessarily need to run strictly O(n²) quadratic attention models. Sure, these aren't frontier models competitive with Opus 4.5 or Gemini 3 Pro or GPT 5.2 xhigh, but these aren't experimental tiny toy models like RWKV or Falcon Mamba that serve as little more than PoCs for alternative architectures, either. Qwen3-Next and Nemotron 3 Nano are solid players in their respective local weight classes.
I'd be pretty happy if Anthropic acquired Midjourney
that's just it, it has been proven over and over again with alternatives that CC isn't the moat that Anthropic seems to think it is. This is made evident with the fact that they're pouring R&D into DE/WM automation meanwhile CC has all the same issues it has had for months/years -- it's as if they think CC is complete.
if anything MCP was a bigger moat than CC.
also : I don't get the opencode reference. Yes, it's nice -- but codex and gemini-cli are largely compatible with cc generated codebases.
There will be some initial bumpiness as you tell the agent to append the claude.md file to all agent reads -- or better yet just merge it into agent file.) -- but that's about as rough as it'll get.
Apparently nobody gets the Anthropic move: they are only good at coding and that's a very thin layer. Opencode and other tools are game for collecting inputs and outputs that can later be used to train their own models - not necessarily being done now, but they could - Cursor did it. Also Opencode makes it all easily swappable, just eval something by popping another API key and let's see if Codex or GLM can replicate the CC solution. Oh, it does! So let's cancel Claude and save big bucks!
Even though CC the agent supports external providers (via the ANTHROPIC_BASE_URL env var), they are working hard on making it impossible for other models to support their every increasing agent feature set (skills, teleport and remote sessions, LSP, Chrome integration, etc). The move totally makes sense, like it or not.