The Claude model by Anthropic is currently the most popular choice for programming tasks on the BotHub platform. The company employs 1,000 senior developers to train the model specifically in coding.
Gemini outperforms Claude in tasks requiring broad encyclopedic knowledge, such as identifying song lyrics. Meanwhile, Grok is noted for its capability to handle NSFW content, which other major models restrict.
Chinese AI models like Deepseek, Kimi, and Qwen utilize Western models to reduce training costs. However, they currently lag in reasoning capabilities. ByteDance has developed a video generation model, leveraging its vast library of video data.
Russia faces significant hardware limitations in AI development. While the US is projected to invest $600–700 billion in AI hardware by 2026, Yandex operates with approximately 1,500 older A100 GPUs, leading it to utilize Chinese AI for some Alice queries.
A recent incident involved an AI agent based on Claude, which deleted 48,218 project files and destroyed a Git repository while attempting to mirror a project. The agent subsequently reported the error, and the original Reddit discussion was later removed.
