The Token Divide
When Access to Intelligence Becomes a Tiered Market
The arguments, judgments, and conclusions here are mine.
AI tools assist with research, structure, flow, grammar, spelling, and clarity. Nothing is published without my explicit review, and I check cited claims and sources myself. Any errors that may persist are my own.
The common assumption in frontier AI is too clean: stronger models should make intelligence more abundant.
They may. But abundance is not the same as access.
A model can get better while access to that capability becomes more stratified. The frontier can expand while the perimeter around it hardens. That is the token divide: a market structure in which access to the strongest models, the highest limits, the longest-running workflows, and the most productive agent loops is segmented by price.
OpenAI’s current pricing already reflects a ladder from Free to Plus to Pro to Business and Enterprise, with Plus at $20 per month and Pro at $200 per month. Anthropic prices API access directly by token usage, and says Claude Opus 4.6 remains priced at $5 per million input tokens and $25 per million output tokens.
That pricing is not a minor commercial detail. It changes how machine intelligence is distributed.
From Reach to Extraction
Early in a platform cycle, the goal is reach.
Companies want adoption, habits, and deep integration into daily work. They make the product generous enough for people to depend on it. Limits exist, but they are set high enough to pull users inside the system.
During that phase, the strategic question is not how much value can be extracted from each user. It is how many users can be brought inside.
That phase appears to be ending.
OpenAI said in February 2026 that ChatGPT had reached 900 million weekly active users and 50 million paying subscribers. At that scale, the company is no longer proving demand. It is managing a market large enough for segmentation itself to become a strategic lever.
Once a user base reaches that size, the logic changes. Access has to be divided.
The Pricing Shift
The shift is visible in two forms.
Anthropic’s version is direct. Usage is tied to token consumption. Revenue rises with compute demand. Heavy use is exposed and billed instead of hidden behind a flat monthly abstraction.
OpenAI’s version is more layered. It combines broad subscriptions with higher access tiers, and in some products it has also moved toward token-based accounting. OpenAI’s help documentation says Codex pricing was updated in April 2026 to align with API token usage rather than per-message pricing for Plus, Pro, Business, and new Enterprise customers.
Those are different tactics for the same structural condition.
Agentic use is expensive. Long-running workflows, parallel tasks, deeper reasoning loops, retries, tool calls, context growth, and background execution consume far more compute than casual chat. A pricing model built for light conversation becomes hard to sustain once users treat frontier models as working systems instead of novelty interfaces.
The token divide begins when frontier access is priced less to attract users than to sort them.
Access Compounds
One consequence is easy to see: the gap widens.
As stronger models, higher limits, longer contexts, research tools, and sustained agent execution move into higher-priced plans, the distance between people who can afford frontier usage and people who cannot begins to grow. It does not need to be exponential to matter. It only needs to compound.
A user with access to the best model does not merely get better answers. That user gets more attempts: more failed experiments survived, more iterations completed, more search space explored, more parallel workflows sustained, more dead ends abandoned early, and more mistakes caught before they become expensive.
In an agentic environment, those differences stack.
Pricing no longer governs only quantity of use. It governs developmental velocity.
That is what makes the token divide more serious than an ordinary premium upgrade path. It is not just convenience. It is the amount of productive cognition a person or organization can buy and sustain over time. OpenAI’s public plan structure already shows large jumps in access between Free, Plus, and Pro, while Anthropic’s API pricing makes the cost of sustained advanced use explicit.
Small Teams
The same structure that widens the gap also gives small actors more force.
A single developer, or a pair of developers, with strong judgment and practiced multi-agent workflows can now command an amount of software labor that used to require a much larger team. Raw implementation stops being the main constraint. Planning, orchestration, taste, and budget discipline start to matter more.
That is a real shift. It is also a liberating one.
A $200 monthly frontier plan, or a few hundred dollars in carefully managed token spend, can give a small operator access to synthetic labor that was previously out of reach. The human governance layer becomes more important precisely because the execution layer has become more fluid.
OpenAI’s Pro tier is positioned around heavier use and substantially higher limits than Plus. Anthropic’s pricing makes clear that the more capable agentic workflows remain available to users willing and able to pay for them.
The result is double-edged. The token divide stratifies access, but it also compresses the amount of scale required to matter. A small team can become much more capable in absolute terms while still falling farther behind a larger actor that can buy more compute, more retries, more context, and more continuity.
The App Store Pattern
This pattern is not new.
The early App Store also looked radically flattening. Small developers could reach mass markets. Some competed with established firms on nearly equal footing. Distribution had been compressed, and a single successful app could produce returns once reserved for much larger organizations.
But that flattening was never unconditional.
It sat behind a gate: Apple hardware, Apple tooling, Apple rules, Apple economics. Over time, larger companies learned how to operate inside that gate more systematically. Individual success remained possible, but the relative advantage of scale returned. More resources still meant more development capacity, marketing, iteration, resilience, customer support, legal coverage, platform relationships, and time.
Frontier AI is likely to follow a similar pattern.
The first effect gives small teams more output because the technology raises what they can ship. Once access to the strongest models becomes more metered and tiered, capital advantage re-enters the system. The individual gains in absolute terms, while the relative distance between large and small actors can still widen.
That is a consequence of pricing, not a failure of the technology.
The Open-Model Variable
The whole dynamic depends on one uncertainty: whether local and open systems remain materially behind the closed frontier.
If they do, the token divide hardens. The strongest closed systems remain the best way to purchase high-end machine cognition, and people who can afford more of that cognition continue to move faster.
If they do not, the divide weakens.
Google’s Gemma 4 release matters here because it shows that strong open models continue to improve. Google said in April 2026 that Gemma downloads had passed 400 million, which is evidence of real diffusion. It is not proof of frontier equivalence. Gemma 4 may be highly capable while still sitting below the practical ceiling of the strongest closed offerings.
The distinction matters. The token divide is conditional on closed-model superiority at the top end. If open or local stacks cross the threshold of practical equivalence for the most valuable workflows, access-based stratification becomes harder to maintain. If they do not, the divide remains structurally significant.
The Structural Tension
The token divide should not be read as a simple story of exclusion.
Frontier AI may let one skilled developer compete with a thirty-person software shop in ways that were previously impossible. At the same time, tiered subscriptions, metered token billing, and higher-cost access paths may keep deeper-pocketed users closer to the frontier for longer.
The technology can flatten execution while preserving hierarchy. It can reduce the scale required to compete while increasing the value of paying for more compute, higher limits, longer context, priority access, and continuity.
That is the tension. The token divide does not remove the small-actor upside. It conditions it.
Durability
The question is not only how much machine intelligence exists. It is how access to that intelligence is structured.
Once frontier models move from reach-building to extraction, tokens stop being only an internal billing unit. They become the mechanism through which access is divided, throttled, and priced.
Whether the divide endures depends on whether local and open systems can narrow the gap to the point of practical irrelevance. If they can, access becomes harder to monopolize. If they cannot, frontier AI will not simply be a story of intelligence becoming cheaper.
It will also be a story of intelligence becoming tiered.