DeepSeek V4 Rumored to Challenge Claude and ChatGPT in Coding
Hangzhou-based AI startup DeepSeek is reportedly preparing to launch its V4 model by mid-February 2025. Insiders suggest V4 is engineered to outperform competitors like Anthropic's Claude and OpenAI's GPT series in complex programming tasks.
The model is expected to excel with long code prompts and aims to surpass Claude Opus 4.5's SWE-bench Verified record of 80.9%. A new training method, "Manifold-Constrained Hyper-Connections" (mHC), may allow DeepSeek to overcome hardware limitations imposed by U.S. export restrictions.
This development could significantly impact decentralized AI and enterprise software, particularly in areas like smart contract auditing and DePIN development.