I’m a happy Microsoft 365 Premium subscriber, which includes access to Microsoft Copilot. But after hearing friends and family rave about Claude, I decided to spend a month using Claude Pro instead of Copilot.
For context, I am a Psychology student specializing in industrial and organizational psychology at the Master’s level. I am blind, and I rely on AI as an accessibility tool to help me understand the world and manage the demands of academic research. Large language models such as Copilot and Claude are especially valuable for summarizing articles so I can quickly judge whether they are worth reading. This helps me avoid the common trap of a strong abstract, introduction, and conclusion that hides an article that is not actually useful.
A standard fifteen page journal article takes a screen reader about two and a half hours to read from start to finish, and since skimming is not possible with blindness, LLMs make the research process far less exhausting.
I prefer tools that are fast, predictable, and free of unnecessary filler. I want the information I ask for and nothing more. I also value knowing the cost of a tool upfront, and more than anything, I need consistent behavior so I can plan ahead without being caught by last minute surprises.
If a tool can outperform Copilot in my workflow I want to know, and that’s what led me straight to Claude.
My Observations
Copilot and Claude market and bill their products in fundamentally different ways. Copilot gives users a set of modes, but it never asks you to choose a specific model. The modes include Smart (OpenAI GPT and GPT Thinking), Think Deeper (OpenAI O class reasoning models), and Search.
Claude takes the opposite approach: it exposes the models themselves. You choose between Opus, Sonnet, and Haiku, and you can pick from multiple versions of each. You can also toggle Adaptive Thinking, which adjusts the amount of reasoning based on prompt complexity.
In short, Microsoft Copilot focuses on the product, while Anthropic Claude focuses on the models and their capabilities. Both approaches can be confusing unless you are a technical user who understands what these terms actually mean.
Copilot offers unmetered usage, while Claude gives you a token allowance, a measurement of how much work the model performs. Once you exceed that allowance, Claude temporarily blocks further usage unless you upgrade your plan or buy additional credits, which can become expensive quickly.
If your workflows are reasoning heavy, such as advanced mathematics, programming, or complex analysis, Claude is extremely capable, but you will pay for that capability.
If you decide to try Claude, it’s important to understand that the reliability of their service leaves a lot to be desired. Their system status page publicly lists every incident, and the volume of outages speaks for itself. During my month using Claude, I frequently ran into problems sending prompts or receiving responses, often having to retry the same request later.
Most of what I used Claude for wasn’t time‑sensitive, and I do keep multiple AI apps on my iPhone, but it’s still frustrating to be unable to use a service I’m actively paying for. The other apps aren’t perfect either, yet none of them seem to experience outages as often as Claude does.
Claude handles context and follow‑up responses far better. It was refreshing not having to take the output from one chat and rebuild it into a new prompt just to continue the conversation which is something I often have to do with Microsoft Copilot.
Claude’s strong reasoning is enabled by default, which speeds up workflows. With Copilot, I can switch from GPT‑class models to O‑series reasoning models, but it requires manually changing modes.
Claude notes that the Opus 4.7 model consumes tokens more quickly. In my own testing outside of programming tasks, I was able to use Opus 4.7 as my default model for research, writing, and even help with understanding math problems without burning through my quota. This was on the Claude Pro plan at twenty dollars per month, not one of the higher‑tier options. It does use more tokens, but in practical day‑to‑day workflows you may not notice the difference. If you are a coder or a scientist who runs heavy reasoning tasks, the increased token usage may affect you more.
Both Claude and Copilot do a solid job with their memory systems, referencing previous conversations and maintaining context over time.
Reflection
AI products tend to work best for me when I stick to a single service at a time. I suspect this is because the service can build a more consistent picture of my preferences and generate more tailored responses. That said, any AI chatbot with a memory system and the ability to reference past conversations would likely offer similar benefits, so it’s not something unique to one platform.
I like Claude and think it’s a strong product. If my daily work involved computer programming or a math‑heavy academic field, I might lean toward it more. But for now, Copilot more than meets my needs.
For now, I’m switching back to Microsoft Copilot, though I’m excited to see Claude’s innovations make their way into other AI tools.
The Honking Goose 