AI Caps as Pricing: How Anthropic's Claude Code and OpenAI's Codex Nudge Upgrades with Usage Limits

AI tools don’t just sell features—they sell access to compute by limits. The way they cap that access shapes behavior, stops abuse, and pushes heavy users to pricier tiers.

By shaping when and how much you can use them, they curb abuse, segment power users, and lift ARPU.

Anthropic (Claude Code): Rolling 5-Hour Windows

Anthropic's Claude Code has 3 plans: $20 plan, $100 plan, and $200 plan.

• Time-boxed usage. Claude Code runs on a rolling 5-hour cap. When you hit it, the meter resets after 5 hours.

• Clear countdown. The app shows when the next window opens, so you can plan work bursts (and even switch as soon as the window resets).

• Mid and top tiers. Upgrades like Max 5× (~$100) and Max 20× (~$200) lift the allowance per 5-hour window. The model stays the same; you’re buying more runway per window.

• Predictable feel. You get steady throughput and fewer surprises: a visible meter, a visible reset.

OpenAI (Codex via ChatGPT): Plan Buckets and Cooldowns

OpenAI's Codex only has 2 plans: $20 plan and $200 plan.

• Tied to ChatGPT tiers. Codex access comes with Plus (~$20) and Pro (~$200)—there’s no $100 middle tier for individuals who want more than $20 but less than $200.

• Soft numbers, hard stops. Limits aren’t spelled out in detail. On Plus, users often report a "User limit reached. Try again in 2 days 21 hours 17 minutes" message and a multi-day cooldown; work resumes later or with an upgrade. The $20 tier is a bait to upgrade to $200 plan.

• Bigger bucket on Pro. Pro generally means a much larger pool and fewer cooldowns, but the jump is straight from $20 to $200.

Here’s the strange contrast: Anthropic sells time (a visible 5-hour window) and lets you buy more window at $100 or $200 while providing strict 5-hour limit.

OpenAI sells tiers (Plus → Pro) and uses cooldowns on the lower tier to create upgrade pressure, with no $100 stop in between while providing generous limits where you can use Codex for 3 days before hitting the upgrade limit message of 3 days.

In short, Anthropic feels predictable and meter-like; OpenAI feels bucketed and scarcity-driven. Both curb abuse and raise revenue, but they nudge users differently: one by showing a clock, the other by hitting a wall after being an extensive user.

Top Tweets of the day

1/

Damn good one. MUST READ!!

2/

TikTok essentially nailed the algorithm for new accounts and it has been a game changer for many creators and businesses.

OTOH, Julia Pintar's TikTok growth hacks are amazing.

3/

Unique meme strategy.

Rabbit Holes

1. We get 34.2M Views a Month. Here's how (just copy me) by Alex Hormozi

2. Six Unconventional Ways to Ship Insanely Fast & Win In Your Market by Enzo Avigo

3. How to Miss By a Mile: An Alternative Look at Uber’s Potential Market Size by Bill Gurley

What’d ya think of today’s newsletter? Hit ‘reply’ and let me know.

Do me a favor and share it in your company's Slack #marketing channel.

First time? Subscribe.

Follow me on X.

More Startup Spells 🪄

1. Roblox's $100 per day cheap Growth Hack to Hit 100,000 Users using Google Ads (LINK)

2. Substack's Leaderboard Ranking Secrets: Growth Insights from #3 Finance Newsletter (LINK)

3. How Starter Story Uses Urgency For Limited-Time Black Friday Sale (LINK)

4. A 3-Page Priming Paywall Flow That 3x'd LTV (Data from a 14 Day A/B Test) (LINK)

5. Pavlovian Conditioning in Marketing: How Brands Use Psychology to Influence Buying (feat. Nokia, Starbucks, Coca-Cola) (LINK)