An analysis published by Forbes about Cursor’s costs has generated viral headlines suggesting Anthropic loses 5000 dollars on every heavy Claude Code Max user. However, a closer examination suggests this figure confuses API pricing with actual infrastructure costs.
The confusion between price and cost
The original article cited internal sources indicating that the 200 dollar monthly Claude Code Max plan can consume up to 5000 dollars in compute. But this figure apparently relies on Anthropic’s API prices, not what it actually costs to serve those tokens.
Anthropic‘s current API pricing for Opus is 5 dollars per million input tokens and 25 dollars per million output tokens. At those prices, a heavy user could indeed rack up 5000 dollars in API equivalent usage. But API price is not Anthropic’s cost.
The OpenRouter reality check
One way to estimate real costs is to look at what competitive providers charge on OpenRouter for similar sized models. Qwen 3.5 397B and Kimi K2.5 are comparable models in architecture and size. These models are offered on OpenRouter at approximately 0.39 dollars per million input tokens and 2.34 dollars for output, roughly 10 times cheaper than Anthropic API prices.
These providers operate profitable businesses, not charities. If real cost is approximately 10 percent of API price, a heavy Claude Code Max user would cost Anthropic close to 500 dollars monthly, not 5000. That represents a loss of 300 dollars on the most extreme users, not 4800.

Most users do not hit the limit
Anthropic has indicated that fewer than 5 percent of subscribers reach weekly limits. For average users, the mathematics likely works out to break even or profit for the company.
The analysis suggests that whoever actually faces 5000 dollar per user costs is Cursor, which must pay Anthropic’s API prices to offer a similar experience. For Cursor, integrating Anthropic models represents a significant economic challenge.
Australian business implications
For organisations evaluating AI development tools, this analysis offers perspective on the real economics of these services. API prices do not reflect infrastructure costs, and competition in the inference market continues pushing margins downward.
Businesses considering building on proprietary model APIs versus open source alternatives should carefully evaluate costs at scale. For a detailed comparison of Claude versus ChatGPT capabilities, see our complete guide to Claude vs ChatGPT.
Considering AI tools for your business? Book a free discovery session with Aivy to understand the real costs.
