Ten points smarter, and the line is somewhere else

For most of the past few years, the pitch for a new AI model was simple: it's smarter, so developers will switch to it. New reporting says that pitch is losing force.

The Information reported last week that, for OpenAI, Anthropic and other leading labs, a 10-point intelligence advantage now earns about 20% more developer traffic. In late 2025 the same lead earned nearly 69%. The outlet's post on X with the figures had passed 240,000 views by Monday.

On Monday morning, Russell Kaplan, president of Cognition, the company behind the Devin coding agent, quoted it: "This is directionally consistent with our internal data." His summary has become the phrase of the day: "Many tasks within software engineering are now 'intelligence saturated.' Once any model can do the task well, cost and speed are all that matter." By mid-afternoon Pacific time his post had about 100,000 views.

The other side. Wharton professor Ethan Mollick pushed back on reading too much into developer numbers. Developers will be "a small part of total token usage over time," he wrote, and "for long-running high value tasks in areas where verification is harder, the benefits for higher 'intelligence' models are quite large."

What the numbers don't say. The full article is paywalled and we haven't read it, so we can't tell you which intelligence score or traffic measure is behind the figures. As TokenPost notes, they describe an association, not proof that smarter models caused the traffic, and they count traffic, not how many developers signed up.

Our take

Our cartoon's joke is simple: the genius model is still a genius, but the line for "rename this variable" goes to whichever model is fast, cheap and right.

That's also good advice. If a task is quick to check, like a rename, a typo, a small test, try the cheaper, faster model first and only move up when it fails. Save the expensive model for long, messy work where a mistake is hard to spot. Your bill will thank you, and so will your wait time.

What we did and didn't do: this story is based on The Information's public summary of its report, the public posts linked below and TokenPost's write-up. We didn't run our own traffic analysis. The figures are The Information's.

MODEL Editors