Stephen Todd
Newsletter
AI 22 April 2026 6 min

Your favourite coding tool is about to get a lot more expensive

Anthropic spent yesterday quietly testing a 5x price hike on Claude Code, got caught, and reverted. The real story is what they admitted in the process.

Yesterday was a bit of a mess for Anthropic, and if you care about the AI coding space at all, you should probably pay attention to how this one played out. Because it tells you everything about where pricing is heading.

The 5x price hike that technically wasn’t

On April 21st, a post went viral claiming Anthropic had pulled Claude Code from the $20 Pro plan. If you wanted it, you’d have to jump to Max at a hundred bucks a month. That’s a 5x price hike with no announcement. Just a sneaky edit to the pricing page. The pricing page was updated so that only Max 5x at $100 and Max 20x at $200 showed the check mark for Claude Code.

And the internet caught fire.

Developer George Pu’s tweet flagging the change hit nearly 900k views within a few hours. A GitHub issue titled “Breaking Change: Claude Code CLI Removed from Pro Plan Without Notice” was filed within hours. Hacker News lit up. Reddit lit up. You know the drill.

Anthropic’s Head of Growth jumped in to clarify, and his version is that it’s “a small test on ~2 percent of new prosumer signups”, and that existing Pro and Max subscribers aren’t affected. A few hours later, they reverted the landing page and docs changes because, understandably, the other 98% of people were pretty confused.

So technically, nothing changed. Technically.

But this wasn’t really about 2%

If it were truly a 2% experiment, they’d do it with a server-side flag. A marketing test. Quiet. Invisible.

Instead, the public pricing pages, the titles of support articles and the entire communication layer were updated as if the change were global. The check marks on the Pro tier got replaced with an “X.” Anyone visiting the site saw Claude Code excluded from Pro, not just the 2%.

And Anthropic straight up admitted the real reason in the same thread. I’ll paraphrase what their Head of Growth said:

When we launched Max a year ago, it didn’t include Claude Code, Cowork didn’t exist, and agents that run for hours weren’t a thing. Since then, we bundled Claude Code into Max and it took off after Opus 4. Engagement per subscriber is way up.

Translation: the $20 tier is bleeding money. People are using Claude Code WAY more than expected. And higher prices are likely coming, they’re just trying to figure out what that looks like.

So the test got reverted. But the intent clearly didn’t.

Why this matters if you’re on Pro

If you’re on Pro right now, you’re safe at the moment. Anthropic promises plenty of notice before anything changes for existing subscribers.

But the math doesn’t really work. Pro users report Claude Code sessions running up to three times longer than on Opus 4.6, often with automated workflows grinding in the background for hours. Some developers are burning hundreds of dollars in tokens a month while paying only twenty. That gap simply isn’t sustainable.

The likely outcome? One of two things.

A new mid tier plan, something like a “Pro Plus” at $40 to $50 a month with limited Claude Code access. Or a straight bump, where Pro gets more expensive, or Claude Code just gets capped hard on the $20 tier.

Either way, the $20 all-you-can-eat era is probably ending. The only question is when.

Meanwhile, over at OpenAI

Here’s where it gets interesting. While Anthropic is tightening limits and testing price hikes, OpenAI is doing the complete opposite with Codex.

Two weeks ago, Sam Altman announced Codex hit 3 million weekly users. And to celebrate, they reset everyone’s usage limits. Altman said OpenAI is resetting usage limits now, then repeating that reset every time Codex adds another million weekly users until it reaches 10 million.

Read that again. Every single million new users, everyone gets their limits wiped. That’s potentially seven more full resets coming down the line. Effectively granting unlimited access through periodic resets until reaching 10 million weekly users.

It’s a brilliant move. Instead of raising prices as usage grows, OpenAI is using growth itself as the reward. Turning a cost problem into a marketing moment.

The strategic difference is wild

Stripping back the noise, here’s what’s actually going on.

Anthropic’s approach: usage is too high, we need to charge more, let’s quietly test a price hike, get caught, backtrack, but signal higher prices are coming anyway.

OpenAI’s approach: usage is too high, great, let’s celebrate it, wipe everyone’s limits, make growth the story, turn users into evangelists.

Now to be fair, these aren’t apples to apples. Codex costs on average $100 to $200 per developer per month in token costs, with a large variance depending on model used, number of instances and usage of fast mode, so OpenAI has its own economic problems here.

But the messaging matters. The perception matters.

Anthropic just spent a day looking like they were trying to sneak a price hike past their users. OpenAI spent that same week looking like they were giving stuff away.

Both companies are probably losing money on coding agents right now. But only one of them is losing trust at the same time.

So what should you actually do

If you’re on Claude Pro and use Claude Code daily, I’d probably lock in an annual plan while you still can. Anthropic’s own messaging makes it pretty clear that Pro as we know it is on borrowed time.

If you’re paying $100 a month or more for Max 5x, you’re fine. That’s exactly the tier they’re funneling power users towards.

And if you’re an indie dev or hobbyist getting priced out, it might be time to seriously try Codex. The limits have never been better, and OpenAI’s growth-based reset strategy means they’re only gonna get more generous before they get worse.

The AI coding wars are heating up. And for the first time in a while, the user friendly player isn’t the one you’d expect.

Catch you in the next one.

Stephen

Sources and further reading

    ↑↓ navigate · ↵ open ⌘K to toggle