Gemini Spark Pricing: Why Google AI Ultra Now Starts at $99.99

The most surprising number out of Google I/O 2026 wasn’t a benchmark or a context window. It was $99.99. That’s the new entry price for Google AI Ultra, down from $249.99 just one day earlier. Google effectively cut the price of their flagship AI plan by 60% to make sure Gemini Spark — the new 24/7 AI agent — had a viable mass-market price point.

Here’s what every Google AI tier actually unlocks in May 2026, why the cut happened, and how the lineup compares to Claude Pro / Max and ChatGPT Plus / Pro on the same shelf.

The new four-tier lineup

Two structural changes underlie all of this. First: Google switched from daily usage caps to a compute-credit budget that refreshes every five hours up to a weekly ceiling. So “limits” mean weekly compute now, not daily message counts — and critically, the compute cost is weighted by prompt complexity, features used, and chat length. A short text question burns far less budget than a long video generation or a complex multi-agent coding task. Heavy users won’t hit limits in proportion to message count; they’ll hit them in proportion to how much actual model work they trigger. Second: each tier sits on the same models — the difference is how much of them you can use, not which ones.

Source: blog.google – Google AI subscriptions from I/O 2026

Why $99.99 (and why now)

For the past year, Google AI Ultra cost $249.99/month. That’s roughly where ChatGPT Pro and Claude Max sit, both at $200. The “premium agentic” tier looked like a small market — power users, prosumers, a few researchers.

Then Google decided to bet on Spark being the consumer hook that pulls a much wider audience into agentic AI. A $249.99 price tag was incompatible with that bet. So Google did three things at once:

1. Renamed the old tier to “AI Ultra Premium” and dropped its price to $200
2. Created a new “AI Ultra” at $99.99 sitting at one-fifth the usage but with Spark included
3. Kept AI Pro at $19.99 as the volume tier for people who don’t need agentic features

The math works because the marginal cost of running Spark for a moderate user is low — most Spark “thinking” is brief polling and triggered actions, not constant Gemini-style chat. Google can afford to bundle Spark into a $99.99 plan because the median user won’t burn the 5× Pro usage budget. The 20× Ultra Premium tier exists for people who actually will.

This puts Google directly in competition with two specific tiers:

• Claude Max 5× at $100/mo (Anthropic’s middle-step tier, also added in April 2026)
• ChatGPT Pro at $100/mo (OpenAI’s middle-step tier, added April 9, 2026)

Google is the third lab to plant a flag at $100. That price point now exists in all three major consumer AI stacks because that’s where the prosumer demand actually is — too much for $20, not justified at $200.

What each Google tier actually gets you

AI Plus — $7.99/mo

Best for: people who want to try Google AI properly without committing to a real plan.

What’s included:

• All Gemini models (Flash, Pro at standard context, Omni Flash)
• 200 GB Google Drive storage
• 200 AI credits/month (used for image and video generation)
• NotebookLM with raised limits on uploaded sources and notebooks
• Daily Brief feature (the new morning-summary feature shipped with I/O)

What’s not:

• Gemini 3.1 Pro at the full 1M context (capped at standard)
• Gemini Spark (Ultra-only)
• 5 TB+ storage
• Priority queueing during peak hours

This is the “I want to play with Omni Flash and have NotebookLM not be annoying about source limits” tier. Real value for the price.

AI Pro — $19.99/mo

Best for: people who use Gemini daily for work but don’t need agentic features.

What’s added on top of Plus:

• Gemini 3.1 Pro at 1M token context — the headline difference. This is what makes long-document analysis, codebase work, and big research projects actually pleasant.
• 5 TB Google Drive storage (so it doubles as your cloud backup)
• 1,000 AI credits/month (5× Plus)
• Priority access during peak hours
• Higher Omni Flash generation quota (specific numbers not published)
• Antigravity 2.0 standard access

What’s not:

• Gemini Spark agent
• Project Mariner browser agent
• 20 TB storage tier
• The 5× compute multiplier on Ultra

This is the tier most professional knowledge workers should be on. It hits the sweet spot of “I get the good model with 1M context” without “I’m paying for an agent I won’t use”.

AI Ultra — $99.99/mo (the new tier)

Best for: people who want Spark and are willing to pay $100/mo for it.

What’s added on top of Pro:

• Gemini Spark beta access — the 24/7 agent. Headline feature.
• 5× the Gemini-app and Antigravity usage of Pro
• Priority Antigravity access (faster queueing for the coding platform)
• 20 TB storage (4× Pro)
• YouTube Premium included (this is worth ~$14/mo on its own)
• Wider rollout in coming countries before lower tiers

What’s not:

• Project Genie (the new video-game-style world model — Ultra Premium only)
• The 20× compute multiplier (that’s Ultra Premium)
• The deepest Spark usage quotas (Spark counts against your weekly budget; heavy Spark users will brush the cap)

Once you account for the YouTube Premium bundle, the effective Ultra price is closer to $85/mo for everything else. That’s the real comparison point with Claude Max 5× and ChatGPT Pro.

AI Ultra Premium — $200/mo (formerly just “Ultra”)

Best for: developers building agentic workflows, researchers, people whose income depends on AI compute.

What’s added on top of Ultra:

• 20× Pro usage budget on Gemini app and Antigravity (4× more than Ultra)
• Project Genie access — Google DeepMind’s world-model research preview
• Project Mariner browser agent
• Highest priority queueing across the board
• Same 20 TB storage as Ultra (no storage bump at this tier)

The $100 jump from Ultra to Ultra Premium gets you 4× more compute, two research-preview tools, and not much else. For most people, the math says stay on Ultra unless you’re hitting the weekly cap repeatedly.

Source: gemini.google – Spark agent overview

How Google compares to Claude and ChatGPT in May 2026

A few honest observations:

Google’s $7.99 AI Plus is the cheapest entry-paid tier. Claude and OpenAI both start at $20. If your AI workload is light and you mostly care about NotebookLM, Drive storage, and basic Gemini access, Plus is the budget pick.

At $20, all three are roughly equivalent for general use. Claude Pro favors writing and coding (and bundles Claude Code at no extra cost). ChatGPT Plus has the broadest ecosystem (custom GPTs, the Apple integration, Atlas browser agent on the desktop app). Gemini AI Pro gives you the longest context window (1M tokens) and the best Google Workspace integration.

At $100, the differentiator is what proactive agent you want. Spark is Google’s: cloud-based, 24/7, Workspace-native. ChatGPT Pro at $100 mostly buys you 5× Codex access and image/video quota. Claude Max 5× is the workhorse tier for coding-heavy users who already live in Claude Code. The choice depends on which ecosystem you’re already in.

At $200, you’re buying compute, not features. All three top tiers are essentially “20× the limits” for the same set of capabilities. If you’re not hitting the limits on the $100 tier, the $200 tier is wasted spend.

The hidden cost gotchas

Three pricing things that don’t show up in the tier comparison table but bite people in the second month:

1. The “compute credit” model means heavy days hit your weekly cap fast. A long Spark-driven research project that ran for six hours could eat half your weekly Pro budget in one day. Google’s new five-hour refresh is friendlier than the old daily cap for normal use, but if your work has spiky days, you’ll feel it. The 5× and 20× multipliers on Ultra and Ultra Premium are largely about insulating you from this.

2. YouTube Premium being bundled with Ultra changes the math. If you were already paying $14/mo for YouTube Premium and considering AI Pro at $19.99, jumping to Ultra at $99.99 is functionally only ~$66 more per month after you cancel YouTube Premium separately. That makes Spark cheaper to justify than the headline price suggests.

3. Antigravity 2.0 is included in AI Pro and up, but “priority access” only at Ultra. If you’re a developer planning to use Antigravity as your primary IDE, the priority queueing on Ultra is the real difference. Pro users get throttled during peak hours. This is mostly a developer concern, but it’s the silent reason a lot of devs upgrade.

What this means for you

If you’re trying out Google AI for the first time: AI Plus at $7.99/mo. Don’t overthink it. Use it for a month, see what you actually do with Gemini, then upgrade if you find yourself hitting limits.

If you’re a knowledge worker using AI daily: AI Pro at $19.99/mo is the right tier. The 1M-token Gemini 3.1 Pro is what makes long-document and long-codebase work actually viable.

If you want to try Spark and have AI replace several hours of your week: AI Ultra at $99.99/mo. Subtract the $14 you save on YouTube Premium if you had it — effective cost is ~$86.

If you’re a developer living in Antigravity or you hit usage caps repeatedly: AI Ultra Premium at $200/mo. The 4× compute on top of Ultra is real and the priority queueing matters.

If you already pay for Claude Pro or ChatGPT Plus: Don’t switch yet. The model differences are real but small, and the friction of moving your work over (custom GPTs, projects, memories) outweighs the upgrade until Spark expands beyond the US.

The bottom line

Google cut Ultra’s price by 60% to put Gemini Spark within reach of the prosumer market. The strategy works if Spark turns into something you genuinely offload work to — and falls flat if it’s another agent that needs more babysitting than it saves time on. The next 90 days of real-user data will tell.

For most professionals reading this, AI Pro at $19.99 is still the right tier unless and until you specifically want Spark. AI Plus is the lowest-friction way to try Gemini seriously. Ultra and Ultra Premium are for people whose work pattern actually needs an agent, not people who want one.

If you want to figure out where Spark fits in your specific workflow before you commit to Ultra, our Gemini Spark for Solopreneurs course walks through the decision frame and the actual setup. First 2 lessons free.

Sources

• Google Keyword blog: Everything new in Google AI subscriptions from I/O 2026
• Fellow AI: Gemini Pricing 2026 — Plans, API & Workspace Cost Guide
• gemini.google: Spark agent overview
• TechTimes: Google Cuts AI Ultra to $100, Launches Gemini Spark Agent at I/O 2026
• Finout: Gemini Pricing in 2026 for Individuals, Orgs & Developers
• The Next Web: OpenAI’s new $100 ChatGPT Pro plan targets Claude Max
• Morph: Claude vs ChatGPT 2026 — Honest Comparison and Real Pricing
• FindSkill.ai: AI Pricing Compared 2026 — ChatGPT vs Claude vs Perplexity vs Gemini