On October 17, Arena, an AI model evaluation platform, revealed that its newly released DeepSeek-V4.1-Flash-1017 (Max) model has claimed the number one position on the Code Arena: WebDev leaderboard. Scoring 1670 points, the model is a mere 12 points behind GPT-6 Sol (Max) and outperforms numerous mainstream closed-source models across various capabilities, especially excelling in software engineering ability assessments. In comparison to its predecessor, the model has seen a roughly 60% reduction in average cost per task, while its generation speed even surpasses that of GPT-6.1's ultra-fast mode. Its API pricing is remarkably low, with overall costs amounting to just one-tenth of those for Kimi K3, successfully positioning it within Code Arena's Pareto frontier and making it an exceptionally cost-effective choice.

DeepSeek-V4.1-Flash-1017 (Max) supports wish-based programming, enabling the one-click generation of games, 3D models, and more, sparking widespread discussion overseas. During practical tests, the model successfully created two functional web projects—a 3D exploration scene of ancient Kaifeng and a web-based racing game—in a single attempt, with a total cost of approximately 6 yuan. It effectively converts textual requirements into interactive works that are both complete and playable, although some details still need refinement. Its high cost-effectiveness significantly lowers the barriers to creative validation and prototype development, speeding up the process of bringing creative ideas to life.