In one reported comparison, Codex was judged the closest visual match when Claude Code, Codex, and Google Antigravity recreated the same Foqos screenshots in React. Claude Code’s version had working toggles, while Antigravity’s activity grid looked more uniform than the reference.
Three agents recreated the same Foqos screens
Foqos, an open-source focus app, was the interface target. Each command-line tool received the same screenshots, and the shared prompt asked it to recreate every pictured screen in React. The agents had no source code or design files, and the prompt prohibited external searches for the original app.
The task prioritized visual fidelity and asked for visible interactions where reasonably possible. It also allowed local or mock data for features that would otherwise require backend or operating-system functions. That makes this a comparison of screenshot-based React interfaces—not native iOS rebuilds or recreations of Foqos’s underlying system functions.
The reported model pairings
The comparison used Claude Code with Opus 5, Codex with GPT-6 Sol, and Google Antigravity with Gemini 3.1 Pro. Higher reasoning was enabled for all three.
| Agent and model | Reported visual result | Distinctive detail |
| Codex — GPT-6 Sol | Judged the closest match, with strong proportions, spacing, cards, element placement, icons, and typography. | Renamed Foqos “Fogos.” |
| Claude Code — Opus 5 | Good overall, but less faithful in details such as icons, spacing, sizing, and typography. | The only recreation reported to have working toggles; it added “Enable Live Activity” and “Strict Mode,” which were absent from the screenshots. |
| Google Antigravity — Gemini 3.1 Pro | Polished and structurally sound, but judged the least faithful of the three. | Its activity grid looked more uniform because it did not preserve subtle lighter-purple distinctions. |
Codex matched the visual reference most closely
Codex was judged closest on the details that shape an interface’s overall appearance: proportions, spacing, card shapes, placement, icons, and typography. But it changed the app’s name from Foqos to “Fogos,” so its visual match did not extend to that piece of text.
Claude Code added working toggles and extra controls
Claude Code was the only recreation reported to have toggles that could be switched on and off. It also included “Enable Live Activity” and “Strict Mode,” neither of which appeared in the reference screenshots. Its result combined working visible interactions with controls that were not part of the pictured interface.
Antigravity’s activity grid looked more uniform
Google Antigravity’s recreation was described as polished and correctly structured, but its activity grid did not retain the subtle lighter-purple shades in the reference. That area therefore appeared more uniform, and the overall result was judged less faithful than the other two.
These are qualitative findings from one comparison of one app and one shared prompt, not a general ranking of coding agents.
