By Mohamed Ali - Technology & Digital Economy Commentary
July 15, 2026 - The AI Duopoly Just Died, and Intelligence Is Now Cheaper Than Your Coffee.
A week ago, if you wanted the smartest AI on the planet, you had two options: Anthropic or OpenAI. That was it. A duopoly. Pay their price or settle for less.
That world is gone.
Look at the latest Artificial Analysis chart, it plots how smart each model is against what it costs to run a single task. The pattern is wild. The frontier isn't two labs anymore.
It's four American labs sitting at the top: Anthropic, OpenAI, and now Meta and SpaceX AI. Meta's Muse Spark and xAI's Grok 4.5 are landing near-frontier scores at around 25 to 30 cents per task, while Claude Fable 5 sits close to $3 per task. Similar intelligence, a tenth of the price. Let that sink in.
Google still isn't on the optimal frontier. Gemini 3.5 Pro is reportedly delayed. If they catch up, we get five American labs competing at the top, and prices drop even faster.
And the Chinese labs? Xiaomi's MiMo and DeepSeek are pushing solid intelligence at 3 to 5 cents per task. GLM 5.2 leads the open-weight frontier, which matters if you want to run a model on your own hardware. Alex Karp said it last week: control your own destiny. Cheap open models make that real for regular companies, not just tech giants.
Related articles
There's another force pushing this along. OpenAI is openly using its big model (Sol) to post-train its small one (Luna). The expensive model teaches the cheap one. That's how frontier intelligence keeps trickling down the price ladder, faster every cycle.
Meanwhile, the top models have a problem money can't fix: compute. Fable 5 keeps getting slower because everyone's piling on. When the best model is sold out, users fan out to cheaper options , and those options are suddenly good enough. Revenue spreads. Competition heats up. Prices fall again.
So what does this mean for anyone building products, payment systems, or remittance tools? The cost of adding intelligence to your product is collapsing. A task that cost dollars last year costs cents today. Fraud checks, document reading, customer support, things that were too expensive to automate are now trivially cheap.
Intelligence used to be scarce. Now it's a commodity in a price war between a dozen labs on three continents.
The smartest thing you can do? Stop asking which model is best. Start asking which one is cheap enough to use everywhere.