Author | Dongmei
In the current era of rapid evolution in AI programming tools, model capability itself is no longer the sole focus of competition. Who controls how the model is used, the pricing structure, and developer channels is becoming the new core of the game.
1 Anthropic Suddenly Blocks Third-Party Tools from Calling Claude, Sparking Strong Community Dissatisfaction
Last night, Anthropic announced it had deployed stricter technical safeguards to prevent third-party tools from "disguising" themselves as official Claude Code clients to bypass rate limits and billing mechanisms for low-cost calls to the underlying Claude model. Additionally, Anthropic was revealed to have cut off access to the Claude model for some competitors, including xAI, with the Cursor IDE becoming a key "trigger point."
So, what exactly happened during this time?
The fuse for the incident was ignited when a large number of developers using open-source code agent tools like OpenCode discovered: their originally functional OpenCode, Cursor, and other tools suddenly could no longer call the Claude model, and some accounts were even directly banned.
Within just a few hours, developers paying $100–200 per month flocked to GitHub to express their dissatisfaction, sparking over 147 upvotes and 245 Hacker News points. Furthermore, users began canceling their subscriptions en masse, describing the forced migration to Anthropic's official tools as "going back to the Stone Age."
The reason for such strong dissatisfaction among developers was the suddenness of the incident. There was no warning, no migration plan, just a sudden lockdown.
This restriction specifically targets OpenCode version 1.1.8 and above. However, GPT-4 authenticated via OAuth continues to work normally. Only the Claude Max functionality has been restricted.
OpenCode is an open-source AI coding assistant that integrates Claude into VS Code, Cursor, and other IDEs. It also adds keyboard shortcuts, context awareness, and multi-file editing capabilities. Developers love it because it integrates Claude's reasoning process into their existing workflows without forcing them to use a terminal.
User @Naomarik posted on a GitHub issue: "If I just use CC (Claude Code) alone, it feels like going back to the Stone Age." He immediately downgraded his $200/month Max subscription and then canceled it entirely. His reasoning: "It cannot meet my workflow needs, nor does it provide the visibility that OpenCode offers."
He is not an isolated case. Another user was in the middle of a project when access was suddenly interrupted. The developer stated: "It was working fine an hour ago, and now this error appears." Another said: "I've been using it all afternoon/evening, and then I ran into this problem."
Multiple users reported abandoning their subscriptions mid-workflow. The reason is simple: paying $200 a month only to use Anthropic's terminal-only Claude Code tool, rather than the IDE integration they actually want.
After this incident, almost simultaneously, another piece of news caused even bigger shockwaves.
Foreign media disclosed that xAI, owned by Elon Musk, has internal developers who can no longer use the Claude model via Cursor.
Initially, this was interpreted as Anthropic's "comprehensive blockade strategy." However, insiders later pointed out that this was actually a separate enforcement based on commercial terms. The key lies in Section D.4 of Anthropic's Terms of Service, which explicitly prohibits two types of behavior: first, using the service to build or train competitive AI systems; second, reverse engineering or copying the service.
xAI engineers were using Claude via Cursor to accelerate the R&D and testing of their own models. In a legal sense, this constitutes "competitive use." Cursor is not the violator here, but it became an amplifier for the violation.
Although this series of cutoffs happened suddenly, looking at the longer timeline reveals that this event was not isolated.
In June 2025, the Windsurf coding environment suddenly had its first-party capacity for Claude 3.x cut off, forcing it to switch to BYOK and promoting Gemini as a replacement.
In August 2025, Anthropic revoked OpenAI's access to the Claude API, on the grounds that the latter used Claude for model benchmarking and security evaluation, violating competition restrictions.
At the time, Anthropic's statement was already very blunt:
"It is not surprising that Claude Code has become the top choice for programmers, and OpenAI engineers are also using it."
But there is a clear line between "using" and "competitive use." xAI is just the latest example of crossing that line.
2 Engineer Explains: It Was Collateral Damage
As discontent in the community continued to ferment, Thariq Shihipar, the Anthropic internal engineer responsible for Claude Code, was the first to step forward to explain.
On X (formerly Twitter), he confirmed that the company had "enhanced protections against Claude Code secure socket spoofing" and admitted that the launch did cause some collateral damage: some users were automatically banned for triggering abuse filtering rules, and Anthropic is rolling back and fixing the related issues.
But this explanation did not quell the dissatisfaction in the community.
The reason is that what was truly cut off was not just "abnormal traffic," but the usage path of an entire category of third-party tools—especially those that use OAuth authorization to leverage a user's personal Claude subscription account to run automated coding agents in external environments.
In other words, the goal of this adjustment was not a few specific bugs, but "the bridge itself."
That demolished "bridge" was the connection between the subscription chat model and automated agents.
Tools represented by OpenCode play a very critical but long-term gray-area role. Their core value lies in: converting subscription-based models originally designed for "human dialogue" into infrastructure that can be called by automated agents.
Technically, such tools often simulate the identity of official clients (like Claude Code CLI), making Anthropic's servers "think" the request comes from an official environment by forging request headers and reusing OAuth Tokens.
This is the so-called "client spoofing."
In the short term, this approach brought great freedom to developers: a fixed monthly fee, no API billing limits, and the ability to run high-intensity agent loops for long periods.
But from the platform's perspective, this bridge has three problems.
First, technical uncontrollability. When errors or performance issues occur inside wrappers like OpenCode or Cursor, the blame often falls on "Claude is unstable" or "the model has gotten worse," yet Anthropic cannot reproduce or diagnose these issues.
Second, distorted usage patterns. The design intent of subscription products is "human-assisted programming," not 24/7 autonomous agents. After third-party tools remove rate limits, the load characteristics of the model undergo a fundamental change.
Third, and most realistically, cost imbalance. On Hacker News, a user used a vivid metaphor to summarize the essence of this conflict:
Anthropic offers an "all-you-can-eat buffet," but the premise is that you eat slowly.
The pricing of Claude Pro/Max (up to $200/month) is essentially designed based on "human interaction rates." The official Claude Code environment uses rate limits and an execution sandbox to ensure this model isn't "eaten out of business."
As Hacker News user dfabulich put it:
"There is a obvious cost gap between Claude Code's $200/month subscription price and Anthropic's token-billed API. In high-frequency usage scenarios, the tokens consumed through Claude Code in a single month, if billed entirely via the API, would easily cost over $1,000."
Because of this price difference, the community widely believes that Claude Code itself is more like an "official pass with special pricing," and Anthropic's true intention is to hope users use the official Claude Code CLI under this subscription, rather than "detouring" through third-party open-source tools like OpenCode.
In this context, the technical solution OpenCode implemented to bypass restrictions was seen by many developers as an "inevitable result": when users have already paid for a $200 subscription, they naturally want to use these capabilities in tools they are familiar with and that offer higher efficiency, rather than being tied to a single official client.
3 Community Reaction: Anger and Understanding
The developer community's first reaction was not friendly.
David Heinemeier Hansson (DHH), creator of Ruby on Rails, bluntly stated on X that this move was "extremely unfriendly to customers." In his view, since users pay for a subscription, they should rightfully have greater freedom of use. He stated:
Confirmed that Anthropic is intentionally blocking OpenCode and any other third-party frameworks, with the paranoid attempt to force developers to use Claude Code. For a company that trains models using our code, our text, and everything of ours, this policy is simply terrible. Please change the terms.
I think it's a good thing for all model providers to launch their own command-line interfaces (CLIs), but honestly, no developer wants to install five different CLIs. They definitely want to learn and use one tool that can control all models. For me, that tool is OpenCode.
This reminds us again why we need open-source software! As a developer, you definitely don't want to be bound by a single model provider. If they feel they can control you, you won't be able to resist extracting their interests. So, try some new models today!
At the same time, I also need to remind everyone that this is not yet settled. As a frontier lab, long-term success depends not only on whether you have the best model right now, but also on how you interact with developers.
On Hacker News, discussions about this matter also presented two voices:
Some developers pointed the finger directly at Anthropic's product strategy, arguing that the company should not create such a huge gap between subscription and API billing. A more reasonable approach would be to design the subscription plan itself as "API point packs," for example, giving a certain discount on top of standard API pricing, rather than providing a de facto "unlimited all-you-can-eat buffet."
But there are also voices standing on Anthropic's side.
Artem K, a developer at Yearn Finance, pointed out that compared to directly banning accounts or retroactively charging API fees, Anthropic choosing a "gentle blocking path" is considered quite restrained.
Meanwhile, the tool ecosystem reacted exceptionally quickly. OpenCode founder Dax Raad lamented on X, "Anthropic's behavior today fully demonstrates why competition is the most important thing in the world."
At the same time, he publicly announced that he will work with OpenAI to strive to make OpenCode work as perfectly as possible with GPT-5, and stated that Codex users can "use their ChatGPT/Pro plans directly in OpenCode."
At the same time, "whether Claude Code should be open-sourced" became another main line of fierce debate.
Those in favor of open-sourcing argue that Claude Code, as a developer tool, derives its core value from ecosystem expansion and community innovation, and long-term closed-sourcing will only spawn more unofficial implementations, ultimately weakening platform control.
Opponents, however, point out that Claude Code is precisely a key asset for Anthropic to establish a differentiated advantage in the programming field, and its importance is even greater than the Claude model itself.
"Asking Anthropic to open-source this tool completely is equivalent to asking the company to voluntarily weaken its own competitive barriers."
Another user analyzed why Anthropic would offer such seemingly favorable prices to subscription users? He stated:
The answer is actually not complicated: this is a classic "vendor lock-in" strategy. What they really want to sell is not the $20/month subscription itself, but the complete ecosystem built around Claude Code.
For most ordinary users, ordinary users will not pay $100 a month just to let a chatbot help them do homework or generate a cake recipe. Under this premise, wanting to make substantial profits from general chatbots is extremely difficult in itself.
But the programming market is a completely different matter. Companies have clearly shown a willingness to pay for AI services that can complete work in extremely short times to replace some manual programmers; and individual programmers themselves are also willing to pay out of pocket for these tools to reduce their workload—even if they know that this trend may weaken their own irreplaceability in the long run. At the same time, companies' requirements for "perfect quality" are not high enough to hinder this substitution. Therefore, programming is widely regarded as one of the few markets in the AI field with truly high profit potential, and various companies are competing fiercely around this direction, while Claude Code is precisely Anthropic's core product form launched for this market.
In this context, simply selling a subscription service without any ecological binding was not Anthropic's goal from the start. This subscription model itself is not profitable, and is not even intended to be profitable immediately, but is more like a typical "loss leader." The real purpose is to make users deeply integrate into the overall ecosystem of Claude Code during long-term use.
In terms of tool experience, the debate is equally intense.
Some users believe that Claude Code still leads in "context management," "tool call stability," and overall DevEx (Developer Experience), especially in terminal and TUI scenarios, where it is clearly better than competitors like Gemini, Codex, and Copilot.
But many developers also state that OpenCode has higher execution efficiency under the same model, takes less time to complete the same task, and has stronger multi-model and multi-vendor switching capabilities; other users prefer tools like Kiro and Q, believing they beat Claude Code in simplicity and stability.
Looking back at this matter now, it seems this is no longer a technical discussion, but a realignment of ecosystem positions.
Reference Links:
Disclaimer: This article is translated and compiled by InfoQ and does not represent the platform's views. Reproduction without permission is prohibited.