A recent visual IQ test involving six AI coding agents revealed that Codex 5.5 emerged as the top performer in terms of method, speed, and cost efficiency. The test, which was not conducted in a laboratory setting, required the agents to tackle 25 visual puzzles on a designated platform, where they were instructed to select the age of 30 and provide a link to their results upon completion.
The results were as follows: - Claude Cowork Opus 4.8 scored 90 in 85 minutes, with an estimated additional 10 points. - Claude Code Opus 4.8 also achieved a score of 90 but took 96 minutes, with about 28 additional points. - Claude Sonnet 4.6 managed a score of 68 in 62 minutes, with no additional points recorded. - Codex 5.5 scored 124 in just 18 minutes, with an estimated additional 12 points. - Codex 5.4 scored 101 in 16 minutes, with approximately 14 additional points. - Codex 5.5, with a different pricing tier, achieved a high score of 131 in 34 minutes, with around 6 extra points.
The scoring was just one aspect of the findings. Codex 5.5's success was attributed to its methodical approach: it carefully collected puzzle images, created organized contact sheets, zoomed in on difficult questions, and reviewed its weaker answers before submission.
It's important to note that the high score of 131 was achieved using a shorter prompt and the site's default age setting, indicating that the conditions were not exactly the same for all agents. Nevertheless, Codex managed to navigate potential obstacles by utilizing Chrome to access the site and complete all 25 answers effectively.
Claude agents, particularly Opus, demonstrated a cautious approach, taking detailed notes and reasoning through each step. However, Codex's more organized and quicker methodology allowed it to excel. The analysis also includes screenshots, examples of failed attempts, and precise prompts used during the test.
The key takeaway from this exercise is that in visual web tasks, a strategic method can outweigh sheer capacity. Despite having a larger context window, Claude was unable to secure a better score, while Codex's two additional minutes translated into a significant 23-point IQ gain.
For more information, visit our website. Your support by subscribing helps us grow, and we appreciate your engagement with our content, which is ad-free.
Informational material. 18+.
