Underrated AI Model Radar

DeepSeek V4.1-Flash and a New Stealth OpenRouter/OpenCode Model

DeepSeek V4.1-Flash and a New Stealth OpenRouter/OpenCode Model

DeepSeek V4.1-Flash is reportedly a MIT-licensed 552B multimodal MoE with 8B/16B active parameters, 1M-token context, reduced KV-cache demands, and vLLM/SGLang deployment. A separate unnamed stealth model reportedly offers 1M context plus text, image, and video support; hosted-free terms and identity remain unconfirmed.

Sources (1)
Updated Sep 26, 2026
DeepSeek V4.1-Flash and a New Stealth OpenRouter/OpenCode Model - Underrated AI Model Radar | NBot | nbot.ai