Recently, news about OpenClaw, often dubbed as 'raising a lobster' in tech circles, has been making waves across the internet. While it has garnered widespread acclaim, there are also voices of dissent, most notably the claim that 'a monthly salary of 20,000 RMB can't even support the operation of a lobster (OpenClaw).' Why is running OpenClaw so expensive?

I. Can a 20,000 RMB Monthly Salary Even Cover OpenClaw Operation?

According to a report from Wall Street See, someone within a social circle is sharing their 'diary of raising a lobster (OpenClaw),' with nearly a thousand people queuing up outside the Tencent Building. Some are even willing to pay 500 RMB for someone to install and set it up at their home, just to get it up and running quickly.

What many people did not anticipate is that the cost of OpenClaw does not lie in the software itself but in the model calls it makes behind the scenes. It is inherently a 'Token black hole.' Every time it performs a task, it consumes a large number of Tokens to interact with the backend large model. Once the task chain is extended, tool calls increase, and memory is enabled, the consumption of Tokens rises rapidly.

A regular chatbot may only require a few hundred Tokens for a single exchange, whereas OpenClaw may require several million Tokens to perform the same task. Some users have reported that searching for information and writing a 2,000-word document can burn through 7 million Tokens; running a simple web crawler test can consume 29 million Tokens; and cases of burning through 50 million Tokens in a single day are not uncommon.

One SaaS company has even provided 'lobster operation subsidies' for all its employees, with regular employees consuming 150 RMB worth of Tokens daily and the technical team consuming as much as 1,000 RMB. More insidiously, OpenClaw has a built-in 'heartbeat mechanism'—even when there is no actual output, the system still automatically consumes about 145 RMB in call fees daily, translating to an average monthly loss of over 5,000 RMB.

Some self-media outlets have even lamented, 'Why do people keep saying lately that a monthly salary of 20,000 RMB can't even cover the costs of running OpenClaw? That's actually true. If you want to use it smoothly, a monthly salary of 20,000 RMB is really not enough to cover the costs; you can only experience it with great difficulty.'

II. Why Is Running OpenClaw So Expensive?

As an open-source software, OpenClaw should theoretically gain wider adoption due to its open-source nature. However, in reality, using it is like raising an expensive pet—even with a substantial salary, it can still feel burdensome. So, what exactly is causing its operational costs to be so high?

Firstly, open-source does not mean free. The accumulation of hidden costs in OpenClaw forms the basic framework of its operational costs. Many users are attracted to OpenClaw's open-source nature, mistakenly believing they can enjoy its powerful features at zero cost, overlooking the core logic of open-source software: open-source means 'code openness,' not 'free usage.' The value of open-source software lies in lowering the development threshold for users, not in eliminating all operational costs. The operational costs of OpenClaw begin to accumulate from the deployment stage and exhibit characteristics of being 'hidden and continuous.'

For individual users, light usage may rely on existing computers for deployment without the need for additional hardware purchases. However, the deployment process requires setting up an appropriate runtime environment, involving the debugging and configuration of various open-source tools. Without specialized technical skills, additional technical service fees may be incurred. Even for enterprise users, to ensure stable operation and concurrency capabilities, dedicated servers, clusters, and even high-performance GPUs need to be procured. These hardware investments can easily reach tens of thousands of RMB, becoming a significant one-time cost.

Additionally, OpenClaw's daily operation relies on various third-party services and plugins. Whether it's speech synthesis, web scraping, or message platform integration, expanding certain core functionalities requires paying corresponding service fees. These hidden costs, although seemingly scattered, accumulate over time and become a significant factor driving up operational costs.

Secondly, the cost of calling large model APIs constitutes the core expenditure of OpenClaw's daily operation. OpenClaw's core capabilities rely on the reasoning abilities of backend large models, and every interaction, decision, and line of code generated consumes Tokens in real-time. According to current media statistics, even for light usage scenarios by individual users, such as daily Q&A assistance, simple file organization, or basic email replies, monthly Token consumption easily falls within the range of 1 to 3 million, equating to approximately 20 to 80 RMB.

This may seem insignificant, but once high-frequency automated usage scenarios are entered, the cost curve rises exponentially. When users utilize OpenClaw for batch file processing, multi-task Agent collaboration, or large-scale web scraping, monthly Token consumption can soar to 3 to 10 million or even higher, with corresponding fees reaching 80 to 300 RMB or more.

For enterprises or heavy individual users who rely on automation for survival, this seemingly insignificant 'data traffic fee,' when scaled up, can easily consume most of the labor substitution dividends. Tokens are no longer abstract technical indicators but the 'oil' of the digital economy era, with price fluctuations directly impacting the viability of automated applications. For most large model companies, this has instead become the most profitable path. Taking the relatively well-known Yuezhi Anmian as an example, it not only exceeded its 2025 annual revenue in just 20 days but also signaled that overseas revenue surpassed domestic revenue for the first time. On the product side, in response to the popular OpenClaw, Yuezhi Anmian quickly launched Kimi Claw, available exclusively to paying users at the 199 RMB level and above, becoming the first domestic 'Five Little Tigers' company to personally enter the cloud Agent product market. This shows just how lucrative the Token business is.

Thirdly, frequent automated tasks inadvertently exacerbate Token consumption. OpenClaw's proud automation capabilities hide a large amount of 'invisible work,' and these silent computational losses are the invisible killers behind high costs. Many users' understanding of costs is often limited to 'what results I see' while overlooking 'what the machine did to achieve those results.' OpenClaw's core value lies in its Agent attributes, enabling automated model disposal, task planning, and self-iteration. However, this is precisely where costs are most likely to spiral out of control.

Imagine OpenClaw performing a seemingly simple task like 'organizing meeting minutes and extracting to-do items.' It does not directly provide the answer. In the background, it needs to first convert speech to text, then call a large model for semantic analysis, followed by formatting, and finally, it may need to self-reflect to verify accuracy. This series of 'thought chain' processes consumes a large number of Tokens at every step. These are the 'background labor' invisible to users and the root cause of soaring costs.

More critically, to ensure the stability of automated tasks, OpenClaw often requires extensive trial and error. Before outputting the final answer, the model may perform multiple internal derivations and self-corrections. For traditional software, running code once incurs a fixed computational cost. However, for an AI Agent like OpenClaw, the same task may result in geometric growth in Token consumption due to contextual understanding biases or randomness.

This uncertainty in costs is unprecedented in traditional software engineering. Users may seem to have only made 'a small step forward' with OpenClaw, but at the underlying logical level, it may have performed countless 'mental calculations,' with each calculation incurring a fee. This 'black box' cost generation mechanism is the fundamental reason why users feel like they've 'spent a lot of money without doing anything significant.'

Fourthly, Tokens are increasingly becoming a structural bottleneck restricting the widespread adoption of OpenClaw. From an economic perspective, the popularity of a technology depends on whether its marginal cost is lower than the labor cost it replaces. Currently, the high price of Tokens is constructing a high threshold. If OpenClaw cannot achieve breakthroughs in algorithmic efficiency or if large model manufacturers cannot significantly reduce API pricing, its application scenarios will be forced to remain limited to high-value, low-frequency domains and will be unable to penetrate into the vast long-tail scenarios.

For the majority of small and medium-sized enterprises and individual developers, if the benefits brought by automation cannot cover the Token consumption costs, then 'running a lobster (OpenClaw)' becomes a luxurious consumption behavior rather than a means of productivity upgrade. This imbalance in cost structure will further widen the technological divide, with only well-capitalized major players able to afford large-scale automated deployments, while ordinary users can only watch from the sidelines.

Therefore, the Token issue is not just a technical optimization problem but also a matter of whether the business model can succeed. If the cost-effectiveness of Token consumption cannot be effectively addressed, the widespread adoption of OpenClaw will inevitably face severe restrictions and may even fall into the awkward situation of 'being well-received but not well-adopted.'

END