As some of you would have seen from my earlier post, the illustration I shared there sums up the answer to this question quite nicely. But the reasoning behind it deserves a deeper look — and that's what this blog is about.

From Free Tier to "I Need More"

I started my journey with Claude's free tier, then tested Sonnet 4.6 (because it was the latest model at that time) and Opus 4.6. At first, I was hesitant to commit, so I took it month-by-month on the $20 Pro plan — and I'm extremely proud to say I consumed the limits to their fullest. In three weeks, I hit 95%+ of both the daily and weekly limits. The model's responses were so good and helped me do so many things that the only limitation I had was the 5-hour utilisation limit cooldown. That was the most frustrating part.

As I got more serious about tinkering with more complex tools and techniques, I needed higher limits — and to get over that restriction, I thought about going to the Max plan.

(Note: This is not sponsored content! I wish it was, but this is just my honest experience.)

The "Pro" Plan Reality Check

My major setback with the $20 Pro plan was hitting the usage limits way too quickly. Part of this was my own learning curve — I didn't yet know how to use the models efficiently. I was using Opus 4.6 for almost anything and everything, which is the most token-consuming and expensive model. While you can often make do with Sonnet 4.6 for general tasks, there's a clear difference in reasoning quality with Opus for complex logic.

I was hitting limits while working on relatively simple things, like my Live AI Timeline. This was all before I knew how to use Claude effectively. Later, I did some research and found a couple of useful techniques. The following video is a good resource for understanding this and can help you save a lot of tokens and utilisation limit:

So then I thought, if I wanted to build complex apps within a short timeframe, I had to upgrade — and I took that leap of curiosity.

Moving to Max: $100 vs. $200

There are two versions of the Max plan:

• $100/month: 5x the limits of the Pro plan.
• $200/month: 20x the limits (essentially for power-users and teams).

I chose the $100 plan. Because obviously. In the first week and a half, I had a lot of time on my hands and I wanted to "Paisa Vasool" — have my money's worth. In the first week, I utilised 90% of my weekly limit in just three days. Even with the Max plan, I was almost hitting daily limits during intense 2–3-hour-long sessions.

The work was significant, the model was handling complex tasks, and the output was high-quality — but it wasn't flawless. Claude still made mistakes, and I found myself switching and experimenting with multiple models and projects.

The First Two Weeks: A Tinkerer's Playground

The first week was great. I'd almost gotten my money's worth and everything was looking good. At this point, I'd also installed Claude Desktop and the Claude app on mobile so I could keep using Claude 24/7 wherever I was. In those first two weeks, I was able to work on:

• An offline tool (medium to high complexity)
• An online tool (low complexity)
• Refinement of 2–3 online tools (low to medium complexity)
• A mobile app (medium complexity)
• Creating a custom AI agent (low complexity)
• 3 web apps (low to medium complexity)
• A couple of smaller research tasks and tools
• Some smaller general chats
• Creating a Figma plugin
• Installing and working with Figma MCP

After Week Two: When Real Life Hits

This is where tragedy struck. I don't know if it was from using — or abusing — too many AI tools, but my PC just gave up on me. Combine that with travel and family events, and I was really struggling to do heavy work with Claude's help.

I had the mobile app, which I kept using, and it was great for quick idea generation and MVP scaffolding. But running entire projects through mobile? That's something I never fully committed to, even though I had GitHub, Vercel, Claude, and Gemini all on my phone. I preferred a PC for deployments and other heavy tasks.

The 1-Million-Token Wall

Across my two weeks of intense use and one week of very light usage, one of the most frustrating discoveries was the context window limit. I was using Opus 4.6 for a specific tool and ran up the 1-million-token context limit for that chat window incredibly fast.

Once that window was full, the chat became unresponsive. I hit a wall. All the project context — the files, the profiles, the logic — was trapped in that window, but I couldn't send another message.

I learned the hard way that prompt length dramatically impacts your token limits and time utilisation. Which also makes one wonder, 60 mins of super focused output vs entire day of low focused outputs. What's more productive?

The Workaround

I discovered you can ask Claude to create a Context File — a summary of the project state. You then download this file and upload it to a new chat window. It helps the new session "remember" the project without re-explaining everything.

The Tricky Bit

You don't know when you'll hit the limit, because you're just focused on getting things done. And when you do hit the limit, Claude sometimes won't even process the command to create the context file. It should be smart enough to auto-generate one at 99% capacity. I eventually got it to work, but it was a struggle.

The Problem with Re-Uploading

There's another issue tied to hitting the context limit. Even with a context file, every time I asked for a logic improvement, Claude would ask me to re-upload specific files. On a Max plan, I didn't mind as much, but on a Pro plan, this constant re-uploading burns through your tokens at an alarming rate. It's not an ideal scenario for efficiency.

After this issue, I found some claims from AI influencers about solutions to increase context windows, but I haven't tried them yet. If you want to understand what a context window is and why it matters, the following video explains the concept very nicely:

Connectivity & Reliability: The Tab-Switching Tax

I also noticed an issue with Sonnet 4.6: during intense logic building, if I moved away from the tab, the connection would often drop. I'd get a "Claude's response could not be completed, please retry" error.

Retrying meant consuming the same tokens all over again for the same task. This forces you to keep the window open and active, preventing you from doing parallel work in other tabs (like Gemini or AI Studio). This meant I couldn't just give a task to Claude, walk away for dinner, and come back to a finished result. (I was expecting that!) I was afraid the task I'd given would be lost. But this was not the case for the Claude Desktop app — more on that below.

Plus, during the few days when Claude had outages, being a $100-per-month subscriber made the downtime feel much more frustrating.

Efficiency Hack: The Gemini + Claude Combo

When I was on the $20 plan, I found a way to be super efficient:

1. I used Google AI Studio (Gemini Pro) to build the foundation and basic solutions for free.
2. I then took that output to Claude Opus and said: "Find all the faults in this and make a list of prompts for Google AI Studio to add these features / fix these issues."
3. I fed those prompts back into Gemini to fix the code.

This saved my Claude tokens for only the most critical "high-brain" tasks.

The Desktop App & The AMD BIOS Struggle

I also tried Claude Desktop to use "Claude Code" and MCP servers. Many influencers talk about this, but few mention the setup hurdles.

If you're a Windows user with an AMD processor, Claude Desktop might not run initially. I had to go into my BIOS to enable Virtualisation (SVM/Hyper-V). Even after that, I had to tweak several Windows settings to get it to work.

But why bother with the Desktop app?

It allows Claude to access a folder directly on your PC. This solves the "re-uploading files" problem and keeps your context cleaner. You can also use Projects to upload relevant documents once, and every new chat in that project starts with that basic context. Installing the desktop app goes a long way toward getting real value out of your Max plan.

What's Still Missing in the Claude Ecosystem

A few features I noticed are currently absent:

• No connection between the online app and the desktop app. If you're using a folder on local storage on your PC and you switch to a laptop from a different location, you can't access those chats or files. They behave like two separate entities, so the experience feels broken between the two touchpoints.
• No control of Claude Desktop via the mobile app. This would be a game-changer for remote tinkering. ( As i was posting this, Claude has come up with Dispatch mode which is basically a solution to this )

Final Thoughts: Is It for You?

Here's the thing about the Max plan: it's not really about the features or the token limits. It's about whether you have enough hours in the day to make those tokens count. The $100 buys you capacity — but capacity without time is just a number on a billing page. Figure out the time part first. The plan will be there when you're ready.

I’m three weeks in, and I’ll soon be sharing separate posts about the offline tools I’m building and the story of how I "burned" a million tokens in just 15 minutes!

Stay tuned, and let's keep building!