The frustration was not intelligence. It was fit.
I recently found myself getting perpetually annoyed with the roughly $20 Claude Pro subscription I had been using for Claude Code, right around the same time I also canceled my Google AI subscription. On paper, neither decision should have been hard. Claude is genuinely strong, and Gemini is strong too. The issue was that the way I work is not the way these subscriptions are trying to optimize by default.
I usually do a real architecture pass before a story goes into development. By the time a coding task lands in the harness, I do not want the model spending a large part of its budget rediscovering the problem framing from scratch. I want it to execute against a deliberate plan. Claude models are often very eager to pull in more context, infer more architecture, and "understand the system" before acting. That is a great property for many users. It is just not always the property I want to pay for on every coding turn.
That difference matters more than it sounds. I have already written about reducing prompt overhead in Dynamic Tool Discovery. Once you start caring about token discipline, you notice very quickly when a coding tool keeps spending budget on work you deliberately already did yourself.
Anthropic's own docs make the tradeoff explicit. Claude Pro is a $20/month plan, Claude Code usage is shared with the rest of Claude, and if you hit the included usage limit you either wait, enable extra usage, or move into pay-as-you-go billing. That is a legitimate product shape. It just turned out to be the wrong shape for my terminal workflow.
Claude Code was the visible pain, but subscription siloing was the deeper one.
What pushed me over the edge was not just hitting limits. It was realizing how little portability I was actually getting from the subscriptions I was paying for. I liked using Pi Coding Harness. I liked using my OpenClaw instance. I liked being able to choose the interface and orchestration layer that matched the job. But increasingly, the models I was paying for inside first-party subscriptions were not the models I could cleanly rely on inside the third-party tools I actually preferred.
That made both subscriptions feel worse. Claude Pro was no longer just "Claude in the terminal." It was also a plan with shared limits that still did not map cleanly onto the harness setup I wanted. Google AI Pro was no longer just "Gemini plus productivity features." It was another bundle that still did not solve my real coding workflow problem.
Google's own positioning around Google AI Pro is broad: Gemini, Gmail and Docs integration, NotebookLM, higher limits in Gemini CLI and Gemini Code Assist, and 2 TB of storage. That is a big bundle. But that was exactly the issue for me. I did not want another wide consumer bundle. I wanted one place where model access mapped cleanly onto the coding workflow I was actually running.
"The question stopped being which subscription had the smartest flagship model. The real question became which subscription let me route different models to different jobs without friction."
Why GitHub Copilot Pro+ fit better
So I canceled both subscriptions and switched to GitHub Copilot Pro+. In practical terms, it felt like replacing two separate ~$20 subscriptions with one $39 seat. The savings and spending basically canceled out. The difference is that Copilot now gives me what the other two subscriptions did not: one broad model bench in a coding product I can actually route through my own preferred harness.
GitHub's current Copilot plan docs are the important part here. Pro+ includes 1,500 premium requests per month and full access to all available models in Copilot Chat. That model list currently includes Claude Opus 4.6, Claude Sonnet 4.6, Gemini 3.1 Pro, and GPT-5.4. That is not a minor convenience. It changes the workflow.
With Pi in front, I can finally use the models for what I think they are best at instead of treating a subscription like a monolithic stack:
Planner: Opus
When I actually want a broad planning pass, strong abstraction, and more patient reasoning, Opus is a great place to spend the bigger thinking budget.
Coder: GPT-5.4
When the plan is already clear and the task is execution, GPT-5.4 feels like the right tool to move code fast.
Reviewer: Sonnet 4.6
For review passes, cleanup, and catching rough edges, Sonnet remains extremely useful. It is the right kind of sharp for that role.
That role split is my workflow, not GitHub's product claim. The important point is that Copilot Pro+ finally gives me one place where the model bench is wide enough for that workflow to be easy instead of awkward.
This is what I actually wanted from a coding subscription
I do not want my subscription deciding that one vendor's UI is my whole workflow. I want the freedom to choose interfaces, harnesses, and review structure while still getting access to the models that are currently strongest for different parts of the loop.
That is why this switch feels more important than "I changed chat apps." It is really a change in how I think about AI tooling subscriptions. I do not want a subscription that mainly gives me one product. I want a subscription that gives me composable leverage.
For many people, the Claude model of "let the assistant decide more of the architecture on the fly" is probably the right answer. I do not even think Anthropic is wrong to build toward that. It is just not how I want to spend coding budget after I have already done the architecture work myself.
Right now, GitHub Copilot Pro+ feels like the better fit. One seat. Broad model access. Higher limits. Cleaner role specialization. Less frustration. That does not make it the permanent answer. But as of March 18, 2026, it is the setup that feels most aligned with how I actually build.
Sources and further reading
- AnthropicWhat is the Pro plan? — Claude Pro is $20/month in the U.S. and includes Claude Code access.
- AnthropicUsing Claude Code with your Pro or Max plan — Claude and Claude Code usage limits are shared, and users can wait, enable extra usage, or move to pay-as-you-go after hitting limits.
- GoogleGoogle AI Plans — Google AI Premium has been renamed Google AI Pro; the plan includes Gemini, NotebookLM, Gemini CLI, Gemini Code Assist, and 2 TB of storage.
- GitHubPlans for GitHub Copilot — Copilot Pro+ is listed at $39/month with 1,500 premium requests and access to all available models in Copilot Chat.