From Copilot to Claude: Same Models, Different Billing

On June 1, GitHub Copilot moved from flat-rate subscription pricing to usage-based billing. Every plan now includes a monthly allotment of AI Credits, with overages on top. Code completions stayed unlimited. Everything else: chat, agentic workflows, code review. All now on the meter.

We panicked on June 2 when just two days in, one developer had consumed nearly the entire team's monthly budget! We did exactly what everyone else did, opened the usage dashboard, forecast our estimated monthly cost, and then sat in shock at how much that cost line had grown.

At that usage, we were burning roughly $60 per dev per day. We'd been paying $21 per dev per month. That's a different category of cost entirely.

We Saw the Logic, Not the Timeline

When GitHub announced the change in late April, we posted about it on LinkedIn — inevitable, the flat-rate era was subsidised, the market was maturing. We understood the logic, but not the timeline, and not how much it would hurt. Understanding the structure isn't the same as being prepared.

We'd written a month earlier about token-based billing as a SaaS COGS problem — focused on AI spend inside the product itself, where untracked token costs sit directly in your margin. Developer tooling is a different line, but the same logic applies: the flat-rate era was always a subsidy, the floor was always going to move, and teams with no visibility into their AI spend were going to feel it first.

What We Actually Did

One of us had been trialling Claude Code for about a month before the billing switch to compare capability. For the simple reason that Copilot felt like it was playing catch-up and we wanted better visibility in the platform differences.

The April announcement made it a team conversation. The June 2 numbers made it an easy one. We moved main development workflows to Claude Code and capped Copilot with a small monthly overage limit.

The Billing Architecture Is the Actual Decision

The underlying models are increasingly interchangeable. The decision wasn't about model quality, but about billing architecture.

Copilot converts your subscription into AI Credits. When it's gone, you're then billed for any overages. This is the budget we wiped out in 2 days. At the time of writing, Claude provides you with an allowance that resets every five hours. This maps naturally to how our development sessions tend to work. You do some targeted focused work, and then naturally adjust to other lighter tasks. Pre-purchased credits carry a 30% discount on top $1,000 becomes $700 in effective spend.

One gives you a runway. The other gives you a rhythm. How long this stays true? We don't know. It's not that Claude is better, but that the billing architecture is more cost effecting and fits how we work right now.

The Unintended Consequence Worth Talking About

There's a subtler problem with usage-based billing that deserves more airtime: it tempts teams to optimise for the wrong thing.

When cost becomes visible at the session level, the instinct is to reach for the cheaper model, shorten the prompt, or avoid the agentic pass altogether. Each of those feels like a reasonable response to a cost signal. But if the cheaper model isn't right for the task, or the shortened prompt loses critical context, you've just traded a small saving for a worse outcome. You're optimising for tokens rather than results.

Model selection does matter — not every task warrants the most capable model, and being deliberate about that is worth the effort. But the bigger discipline is context hygiene. Keep prompts short and specific. Sanitise context before feeding it in. Avoid long, iterative conversations with an LLM where earlier turns are padding the window without adding value.

Usage-based billing will change developer behaviour. The question is whether that change pushes teams toward more cost efficient practices, or just toward the cheapest ones.

We're Not Married to This Decision

Worth saying plainly: if anyone restructures their pricing again, then we'll have to consider another move. We're fully embedded in the GitHub and Azure ecosystem and the rest of our stack hasn't changed.

The models are converging, the tooling is increasingly portable, and the billing structures are only going to continue to evolve. The best position to be in, is to have the ability to quickly evaluate these decisions, and be willing to change when the merits do. In a landscape that shifts almost every week, agility is the only strategy that holds.

How did your team handle the repricing? Are you staying, moving, or still staring at the dashboard? Tweet us on X with #intutobuild and tag us @intutohq.

Authored by Aaron Leggett, Principal Product Architect at Intuto. Photo by Gratisography.