DeepSeek-v4 has a comical 384K max output capability

DeepSeek-v4 has introduced a staggering 384,000 token maximum output capability, dwarfing the typical 4,000-8,000 token limits of most large language models. This allows the model to generate extremely long, coherent outputs without truncation. A Reddit user, zsydeepsky, tested this by asking the model to create a comprehensive single-HTML web operating system. The model responded by generating a 100KB HTML file that functions as a basic OS interface, complete with a desktop environment, file manager, and applications—all within a single HTML file. The user described being "speechless" at the result.

The implications are significant for developers and content creators. DeepSeek-v4 can now generate entire codebases, long-form documentation, or complex reports in a single request. This reduces the need for multi-step generation and manual stitching of outputs. While the quality of the generated code may vary, the sheer volume and coherence represent a leap forward. The model's ability to produce a functional 100KB web OS from a single prompt demonstrates its potential for rapid prototyping and automated software generation. However, users should verify outputs for security and correctness before deployment.