In a recent LinkedIn post, Ruben Hassid offers practical strategies for users of the AI model Claude to manage their subscription costs and avoid hitting usage limits prematurely. Hassid, a proponent of efficient AI utilization, details how Claude’s credit system can lead to unexpected expenses if not understood properly.
Hassid highlights a core misunderstanding of how Claude processes information, stating:
“It re-reads the entire chat every time you send a prompt. Prompt 30 costs about 31x more than message 1.”
This fundamental aspect, according to Hassid, means that the cost of interaction escalates significantly with each subsequent prompt within a single chat session. He further illustrates potential cost pitfalls, such as uploading the same document multiple times across different chats, which incurs redundant charges, and using more powerful models like Opus for simpler tasks that a less resource-intensive model could handle.
Understanding Claude’s Credit Calculation
Ruben Hassid’s analysis centers on the sequential nature of Claude’s processing. As he explains, the AI’s architecture requires it to re-evaluate the entire conversation history with each new input. This means that the token count, and thus the cost, is cumulative and multiplicative. “Prompt 30 costs about 31x more than message 1,” Hassid warns, emphasizing the exponential cost increase.
He points out common user behaviors that exacerbate this issue. For instance, repeatedly uploading the same reference material across different chat sessions leads to paying for the same data processing multiple times. Similarly, employing the high-tier Opus model for tasks like basic grammar checking, which could be handled by the more economical Haiku model, results in unnecessary expenditure.
Hassid’s Nine ‘Stupid Simple’ Fixes for Efficient Claude Use
To combat these inefficiencies, Hassid outlines nine actionable tips designed for daily implementation:
File Optimization
Hassid’s first fix involves significantly reducing file size before uploading. He suggests copying PDF text into a Google Doc, then downloading it as a Markdown (.md) file. This method can drastically cut down token usage, transforming a 15-page PDF that might cost around 45,000 tokens into a .md file requiring approximately 2,000 tokens.
Strategic Chat Planning
For complex tasks like financial modeling, Hassid recommends a two-stage approach. First, use Claude Chat to outline the plan and ask clarifying questions. Once the plan is finalized, paste it into Claude’s Cowork feature with the instruction to “Build exactly this.” This separates the planning and execution phases, potentially optimizing resource use.
Leveraging AI Prompts for Efficiency
Instead of users crafting lengthy prompts, Hassid proposes letting Claude initiate the interaction. He notes that a 500-word prompt consumes 500 tokens, whereas a click or a simple AI-generated question costs almost nothing. His suggested prompt structure is: “I want to [TASK] to [SUCCESS CRITERIA]. Ask me questions using AskUserQuestion before you start.”
Utilizing the Edit Function
Hassid points out that correcting a mistake is more efficient by editing the last sent prompt. “If you want to correct a mistake, go to your sent prompt, click ‘Edit’ on your message, retype your prompt, and hit Save,” he advises. This allows Claude to regenerate from the corrected point, avoiding the cost of a completely new prompt.
Conversation Summarization
To manage long chat histories, Hassid suggests summarizing the conversation every 15 prompts. By prompting Claude to “Summarize this entire conversation into a brief,” users can then start a new chat with this concise summary, effectively reducing a massive token count (e.g., 105,000 tokens) to a minimal one (e.g., 500 tokens).
Model Selection
Choosing the right model is crucial, according to Hassid. He recommends using Sonnet for tasks requiring less than 30 seconds of processing, Opus for execution and multi-step work, and Fable for brainstorming or complex problem-solving. “Pick the model before you start,” he stresses.
Project Folders for Repeatable Tasks
Hassid suggests using Claude’s ‘Projects’ feature for recurring tasks involving the same set of files. Uploading documents to a Project once allows all subsequent chats within that project to access them without re-tokenizing, saving on repeated processing costs.
Disabling Unused Features
Users can optimize costs by disabling features not required for a specific task. Hassid advises turning off tools like Web Search and connectors unless they are essential, as “Every idle feature adds tokens.”
Batching Requests
Finally, Hassid recommends batching multiple requests into a single prompt. Sending “Summarize this, list the key points, and write a headline” as one message is more efficient than sending three separate prompts, as the latter forces Claude to reload the full context multiple times.
Hassid provides further resources on his website for a more comprehensive breakdown of his AI optimization system.
📝 About This Content
This article is based on insights shared by Ruben Hassid on LinkedIn.
📅 Originally posted on September 1, 2026 | View original post on LinkedIn →