A new daily ranking of AI-built games finds Claude Opus behind 54% of new games, while OpenAI-based games earn up to 58% more likes per play.
Developers are building more games with Anthropic's Claude Opus than with any other AI model, but players seem to like games built on OpenAI's models more. That is the main finding of the first Playgama AI Game Index, which looked at more than 4,200 AI-made games and 1.1 million launches.
Claude Opus accounts for 54% of new AI-built games. On the player side, games built on OpenAI's Astra and Sol models earn up to 58% more likes per play. So one model wins on volume, and another wins on how players react.
The index comes from Playgama, a web gaming company. It uses live data from Playgama Sandbox, its platform for AI-built games. Instead of testing code or running lab benchmarks, it tracks what players actually do: playtime, likes and each model's share of published games. It relies on in-game analytics, not surveys or human judges.
Creators can join by publishing to the Sandbox and tagging the model and agent they used. A game counts toward market share from day one. It enters the playtime and like rankings once it reaches 20 plays.
The rankings shift quickly. Playgama founder Dmitry Kachmar said Sol replaced Astra in just two weeks. He also said a model no one used in September leads the rankings in October. The index updates daily, so each change shows up the day after it happens.
Kachmar also argued that the model is not the whole story. In his words, many winners are creators who test faster, adapt faster and shut down weak ideas faster when the data shows little potential.
For anyone building with AI, the lesson is that picking the top model is not a guarantee. AI can speed up making a game, but a game nobody touches is just as easy to produce. Testing, tweaking and dropping ideas that fail appear to count for more.
The data comes from one company's platform, and Playgama also sells the tools around it. The material does not say how the likes-per-play gap was calculated for each model. Rankings may also look different soon, since AI labs keep releasing updates.
The model behind an AI-built game matters less than how quickly its creator tests, adjusts and drops ideas that fail.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
