The new OpenAI GPT-6.2 Astra and GPT-6.2 Sol models showed high results in tests of the DesignArena platform, which evaluates the capabilities of artificial intelligence in game development. Initially, the models took first and third places, respectively, but over the following days, competitors pushed them out of the leading positions. As of November 22, GPT-6.2 Astra is in second place, and Sol is in fourth.

Image source: Mariia Berezovsky/unsplash.com
According to the current DesignArena rating, the leader is the Claude Fable 5.2 model from Anthropic with a score of 1536 points. In turn, GPT-6.2 Astra scored 1502 points, taking second place. Third place went to GPT-6 Astra (max) with 1457 points, and GPT-6.2 Sol was in fourth position with 1437 points. The top five is closed by Claude Opus 5.5 (1435 points).

The GPT-6.2 Astra has a benchmark win rate of 75.4%, compared to the Sol’s 69.6%. For comparison, for the leading Claude Fable 5.2 it reaches 79.2%. Hundreds of users took part in the tests, evaluating the results of generating different models.
DesignArena uses a comparative rating system based on people’s votes. Users are shown the results of two different AI models and asked to choose the most successful option. Based on such comparisons, the Elo rating is calculated. The platform also takes into account the speed of task completion and the cost of generation.

According to initial results, GPT-6.2 Astra in High mode creates an average game project in 6.8 minutes at a cost of about $0.98 per generation. GPT-6.2 Sol copes with a similar task in approximately 4.6 minutes, and the generation cost is only $0.22. Thus, the junior model costs about four times less.
Particularly noteworthy is the comparison of GPT-6.2 Sol with the previous leader in the rating – GPT-6 Astra (max). According to DesignArena, the new model provides comparable quality results, but is 40% faster and costs about four times less.
However, DesignArena’s results should not be taken as an objective assessment of the quality of the program code or the performance of finished games. The platform primarily measures user preferences, and the final positions depend on which generation results they consider more successful.
It is noteworthy that the success of OpenAI models is not limited to game development. In a separate DesignArena rating dedicated to the creation of web applications in HTML format, GPT-6.2 Sol takes first place with 1403 points, and GPT-6.2 Astra is in third place with 1391 points.
If you notice an error, select it with the mouse and press CTRL+ENTER.


