• Vercel finds a typical AI model peaks within four weeks of its launch
  • Anthropic is outperforming OpenAI in attracting new teams of users
  • Users are keen to stay ahead of the curve, with no show of loyalty

The time spent with a new AI model is decreasing, with teams using paid AI services sticking with a new model for an average of four weeks. At that point, they switch to another new model, aiming to keep their AI code development or agents competitive.

Vercel’s AI Gateway Production Index report for November 2026 claims this trend indicates that AI models effectively have a half-life of around six weeks. At that point, attention and use of the model decreases as teams and developers proceed to switch to the latest shiny alternative.

Illustrating the trend, Vercel reports that Anthropic Claude Opus 5.5 was the top model by spend in its first full month, following its launch on September 22, 2026.

The AI half-life

While the concept of an AI model having a half-life may initially seem outlandish, it actually makes perfect sense.

Vercel’s analysis of token transactions via its gateway shows that teams adopt new models within days of them being launched. Assuming the use of a code editor or tool wrapper, pointing a development environment to a new model can be achieved relatively painlessly.

Vercel’s data breaks down the adoption rate into 00-14 days, 15-30 days, and 31-60 days, and highlights a trend towards increased early adoption from June 2026 onward. By September, almost 17% were using models that had been available for under two weeks.

But older models are retained to some degree, but these are typically relying on less usage and, consequently, token spend.

So, with new models being released relatively regularly by competing AI platforms, and with easy model switching essentially baked into developers environments, they’re being used with an initial intensity that wanes when the next model launches. Hence Vercel’s determination that models “have a half-life of about six weeks.”

Video models

Vercel also determined that the top model by spend in its first full month in October was Anthropic Claude Opus 5.5, drawing just over 14% in spend.

Meanwhile, the most widely used paid models are dominated by Anthropic. Sonnet 5.5 (26%), Haiku 5.5 (22%), Haiku 4.5 (20%), and Opus 5.5 (18%) are the most widely-used, although there is duplication, with many team apparently relying on multiple models.

Significantly, image and video generation was also prominent, with more spend and adoption than in previous months. One in six AI development teams are now using a video model, up from 10% in previous months. This is where OpenAI had a rare chance to shine in Vercel's report, with its GPT-Image-2.5 Sunburst leading in image generation model spend during October.