AICodeKing reviews the API release of DeepSeek V4 Pro 0813 and distinguishes official or circulating claims from confirmed details. The video reports strong agentic benchmarkA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions. scores and unusually low token pricesToken pricing is the rate an AI provider charges for processing input tokens, generating output tokens, or reading cached tokens., but cautions that rumored parameter counts and context length were not clearly documented by DeepSeek at the time of recording.
AICodeKing evaluates DeepSeek V4 Pro 0813 on eight Kingbench tasks: an elevator simulation, interactive contact-lens case, folding-table model, SVG illustration, bow-and-arrow game, permutation problem, autonomous local fine-tuning workflow and dual-time-zone wristwatch. The scores total 61 out of 80, or 76.25 percent.
AICodeKing reports strong mathematical reasoning, long-horizon execution, planning, front-end workFront-end code generation uses AI to create or modify the user-interface code of websites and applications from instructions or examples. and three-dimensional interfaces, but also identifies meaningful weaknesses. On simple tasks, the model can overthink and reach a worse answer than a smaller variant, while straightforward fixes can trigger unrequested abstractions and broad edits.
Watch on YouTube
