Google splits Gemini into three tiers, ditches prompt caps

What happened

Per The Decoder, Google used I/O 2026 to overhaul Gemini access into three subscription tiers priced from $7.99 to $99.99 per month, retire daily prompt limits, and shift to a consumption-based compute model. The reorganisation also introduces new models including Gemini Omni and an agent called Gemini Spark.

The pricing structure matters less than the architectural shift underneath it. Google is now metering AI by compute consumed, not by prompts submitted. That puts Gemini in line with Anthropic and OpenAI, both of which have moved similar directions over the past year. The Decoder frames this as an industry-wide trend, and it is.

The headline numbers: three tiers, a 12x price spread between entry and top, and a stated end to fixed daily caps. For brands trying to understand where Gemini's answer surface is heading, this is the more interesting signal than the dollar figures.

Why it matters for your brand

Compute-based pricing changes who uses which model for what. At $7.99, users get rationed access to the cheaper, faster Gemini variants. At $99.99, they get extended runs on the heavyweight reasoning models and agent workflows. That stratification matters because the model serving the answer determines which sources get cited. A Flash-tier model and an Omni-tier model do not retrieve, weight, or attribute the same way. Your brand's visibility in Gemini is now a function of which tier the user is on.

For financial services brands, this is consequential. A retail banking customer asking Gemini about mortgage products on the entry tier is getting a different citation set than a wealth advisor running deep research on the top tier. If your content strategy has been optimised for "Gemini" as a monolith, you are optimising for an audience that no longer exists. The practical move: map your buyer personas to likely subscription tiers and test prompts against the model variants those tiers expose. Enterprise procurement teams almost certainly sit on the top tier. Mass-market retail users do not.

For multilateral institutions and policy bodies, the agent layer (Gemini Spark) is the development to watch. Agents do not just answer questions; they execute multi-step research, pulling from sources iteratively. An agent researching climate finance or disaster risk frameworks will hit your published reports differently than a single-shot prompt does. If your PDFs are not machine-legible, your structured data is thin, or your canonical URLs are unstable, agents will route around you. UNDRR, CGAP, and similar bodies have authoritative content that agents should be citing by default. Whether they actually do depends on retrieval-friendliness, not authority alone.

For major industrial groups, the consumption model creates a new variable in B2B buyer journeys. Procurement researchers running agentic workflows on the top tier will burn compute on long-context tasks: comparing sustainability disclosures, parsing technical specifications, evaluating supplier ESG claims. Brands whose content is dense, well-structured, and citation-ready will be pulled into those workflows. Brands relying on PDF brochures and gated whitepapers will be invisible to them. Holcim-style industrials should be auditing how their technical documentation renders in agent retrieval, not just in human-readable web pages.

For philanthropic and policy institutions, the implication is distributional. Removing prompt caps means heavy users (researchers, journalists, analysts) will run more queries, deeper queries, and more comparative queries. Your think-tank report is now competing for citation slots in queries that would not have been asked under the old caps. That expands the citation surface but also intensifies the competition for it. Volume of model output goes up; share of voice becomes harder to hold.

The signal in context

The industry is converging on compute-based pricing because daily prompt limits were always a proxy for cost, not a feature. OpenAI, Anthropic, and now Google have all moved toward usage-metered consumption in some form. The strategic implication for brands is that the gap between casual AI users and power AI users is widening, and power users are disproportionately the ones making B2B buying decisions, writing policy briefs, and drafting analyst reports. Optimising for the median Gemini user means optimising for the wrong audience if your buyers are on enterprise tiers.

The second contextual point: agents are now a standard layer in the consumer subscription, not a separate enterprise product. Gemini Spark sitting inside the consumer tier structure means agentic retrieval is no longer a future-state concern for marketing teams. It is a current-state reality. The brands that have invested in structured data, clean canonical URLs, and machine-legible reference content will benefit. The brands that have invested in glossy campaign sites will not. That divergence will become visible in citation share within two to three quarters.