Leaked: DeepSeek V4.1 Flash open weights, claims 86x cheaper and 420+ tok/s

ns123abc · x · 2026-09-10

An unverified post claims DeepSeek dropped V4.1 Flash with fully open weights, beating GPT 5.6 Sol, Opus 5 and every Chinese model on coding and cybersecurity at 86x lower cost per million tokens, running 420-507 tok/s (faster than Gemini 3.8 Flash), described as the "smallest model in our new architecture family." None of this is officially confirmed — treat as a rumor pending DeepSeek's announcement.

Related event: DeepSeek Releases V4.1 Flash: New Architecture, Native Vision, MIT Open Weights(35 posts)→

Original post →

More from Models

Models channel →