Google cut AI Ultra to $100 at I/O. Now all three AI giants charge the exact same $20, $100, $200.

What Google actually changed (the headlines got it half right)

A lot of the I/O coverage ran with "Google cut AI Ultra from $250 to $100." That is not quite what happened, and the difference matters if you are deciding what to pay. Google split Ultra into two steps. The old $250 plan, the one with the highest limits, dropped to $200 and kept its 20x usage allowance. Below it, Google created a brand new Ultra tier at $100 with 5x the limits of AI Pro. So Ultra now starts at $100, but the plan that used to cost $250 became a $200 plan, not a $100 one.

Why bother with the distinction? Because the $100 Ultra is a genuinely new product, not a discount on an old one. It is Google's answer to a gap it did not have a clean response for. Anthropic had Max 5x at $100 and Max 20x at $200. OpenAI matched that ladder when it launched a $100 ChatGPT Pro tier on April 9 and kept the $200 one above it. Google was the outlier, selling a single $250 Ultra plan with nothing between it and the $20 AI Pro tier. The I/O move closes that hole and lines Google up dollar for dollar with its two biggest rivals.

AI Pro stays at $19.99, and the cheaper AI Plus tier holds at $7.99. The whole consumer stack now reads $8, $20, $100, $200, with the top three steps mirroring what you see at OpenAI and Anthropic almost exactly.

What each $100 plan gets you

Same price, very different boxes. The $100 tier is where the three companies stop looking identical and start showing what they think a power user actually wants.

Google is buying the comparison with bundle weight. A full YouTube Premium membership runs about $14 a month on its own, and 20 TB of Google One storage is normally a $100-a-month line item by itself. Stack those against the AI features and the $100 Ultra plan is doing a lot of work outside the chat box. OpenAI and Anthropic are betting the other way: fewer bundled goodies, more raw model and agent capacity for people who came for the AI and nothing else.

The catch nobody puts on the pricing page

Matching prices is the easy part. The harder shift, and the one that affects your actual experience, is that all three are quietly tightening what "5x" and "20x" mean. Google's I/O announcement came with a new compute-based usage model: paid plans refresh their allowance every five hours against a weekly cap, with pay-as-you-go top-up credits once you run dry. That is the same metered shape OpenAI uses for Codex and the same direction Anthropic is heading.

Anthropic has the most concrete version of this. On June 15, 2026, Claude splits the subscription into two pools and moves automated tool usage, the Claude Agent SDK,claude -p, and third-party harnesses onto a separate monthly credit billed at full API rates. The credit equals your plan price and does not roll over. We covered the mechanics in detail in our Claude Agent SDK billing breakdown. Interactive Claude Code in your terminal is untouched, but if you bought Max 5x to run the Agent SDK or headless claude -p jobs all day, your June bill is going to behave differently than your May one.

So the sticker prices converged, but the all-you-can-eat era they used to imply is closing. Some analysts argue the subsidized unlimited tier is already over, replaced by compute you meter out. The $100 plan is no longer a flat ticket to unlimited use. It is a budget of compute that refills on a schedule, and the three companies are converging on metering as fast as they converged on price.

$100 a month, or $100 of tokens?

Here is the question this site exists to answer. A $100 subscription and $100 of raw API spend are two different products. The subscription gives you an app, bundled extras, and a usage allowance that resets on a clock. The API gives you tokens, full stop, with no app and no floor. So how far does $100 of tokens go on each flagship? Assuming a 75 percent input, 25 percent output mix by volume, which is typical once you account for chat history and retrieved context:

For a person sitting in a chat window, those numbers say buy the subscription. A heavy daily chat habit burns a few million tokens a month, well inside what $100 of Opus or GPT-5.5 would cover, and the subscription throws in storage, video, and an app on top. You would have to be a genuinely extreme individual user to spend $100 of API tokens through a chat interface alone.

The math flips the moment you go programmatic. An agent looping over a codebase, a batch pipeline, or anything calling the model without a human in the seat will blow past a consumer allowance fast, and that is exactly where the subscription's five-hour refresh and weekly cap start to bite. There the API wins, because you can route the cheap tasks to Gemini 3.5 Flash at a third of the blended cost of Opus and only reach for the expensive model when the task earns it. Run the numbers for your own request shape in the cost calculator before you assume either side is cheaper.

So which $100 plan should you buy?

Since the price is now a constant, pick on what you do, not on what it costs. The honest split:

• Claude Max 5x if you live in a coding agent. Opus 4.7 still leads SWE-bench Verified by a wide margin, and Claude Code on a Max plan is the most productive $100 in this bracket for a developer, at least until the June 15 credit split changes the heavy-usage math.
• ChatGPT Pro if you want the widest single-app toolkit. GPT-5.5, Codex, Sora video, and Deep Research under one login is the most complete "do everything" AI plan, and the Codex 10x promo running through May 31 sweetens it for now. Step up to the $200 tier if you specifically need GPT-5.5 Pro or the 1M-token context.
• Google AI Ultra if you want the most total value for the dollar. The 20 TB of storage and full YouTube Premium membership are worth real money on their own, and Gemini Omni plus the Spark agent round it out. For a non-developer who also wants their cloud storage and media sorted, it is the easiest $100 to justify.

What you should not do is pick on price, because there is no price difference left to pick on. The three giants spent two years racing each other down to the same three numbers. The competition has moved off the rate card and onto what each company can stuff behind it.

Sources

• Google: AI subscriptions update from I/O 2026 - new $100 Ultra tier, $200 top tier, compute-based usage model, plan inclusions
• Engadget: Google AI Ultra now starts at $100 a month - tier restructure detail and pricing breakdown
• Anthropic: Claude plans and pricing - Claude Max 5x at $100, Max 20x at $200, model access
• OpenAI: ChatGPT plans and pricing - ChatGPT Pro $100 and $200 tiers, included features
• Axios: How Google plans to win the AI war - the $20/$100/$200 ladder convergence and price-war framing