GPT-5.6, Grok 4.5, Claude, and Muse Spark build the same 4 apps

# Summary Twelve AI models were tested building the same four applications to compare their coding capabilities, including OpenAI's new GPT-5.6 (in three tiers), Meta's Muse Spark 1.1, Grok 4.5, Claude variants, and several open-source models. The results show GPT-5.6 Sol and Luna performed best on the first task (a Doom-style raycaster maze), with GPT-5.6 Sol achieving 5/5 playable builds, while open-source models like DeepSeek and Qwen showed more mixed results. The researchers published all raw attempts so readers can independently evaluate performance rather than treating the results as a definitive scientific benchmark.