DeepSeek V4 Coding Model Previewed with Novel "Engram" Architecture, Mid-February Launch Expected

Summary

Reports emerged in late January 2026 that Chinese AI lab DeepSeek is preparing to release its V4 flagship model around mid-February 2026, likely coinciding with the Lunar New Year (February 17). The Information reported on January 9 that the model targets coding dominance, and by late January leaked benchmarks and architectural details had surfaced, generating significant industry discussion. DeepSeek V4 introduces a novel "Engram" conditional memory architecture (published January 13, 2026) th