GPT-5.6, Grok 4.5, Claude, and Muse Spark build the same 4 apps (www.tryai.dev)

In a recent performance comparison of AI models, GPT-5.6, Grok 4.5, Claude, and Meta's Muse Spark 1.1 tackled four challenging apps: a raycaster, a Rubik's cube, a calculator, and Conway’s Game of Life. Each model participated in multiple attempts, generating a diverse set of results. The tests revealed that while GPT-5.6 showed strong performance in several areas, including the raycaster, Claude Fable 5 excelled particularly in solving the Rubik's cube. Despite some models falling short, Muse Spark made a surprising impact, showing itself as a competitive entry in the coding model landscape. This build-off is significant for the AI/ML community as it highlights the nuanced capabilities and drawbacks of various models, particularly emphasizing the gap between state-of-the-art (SOTA) models and open-weight alternatives. Key technical insights revealed that while GPT-5.6 performed well in complex tasks, simpler tasks like Conway's Game of Life were effectively handled by models like Qwen 3.7 and GLM-5.2 at a lower cost. Muse Spark emerged as a promising option, positioning itself between Grok and open-weight models. These findings encourage developers to thoughtfully select models based on task complexity and cost-effectiveness, underscoring the notion that higher investment does not always guarantee superior results.