1 comment

[ 0.27 ms ] story [ 11.2 ms ] thread
It's amazing to make it go through a whole book with little hallucination. It still lacks multimodal capabilities though.

I hope they open-source DeepSeek v4 in Q1 2026 with enhancements to coding, agentic, and multimodal capabilities. Usage limits and timeouts are the pain points of all other models.