deepseek-v4.1POST
User take: V4.1 Flash at 763B is no flash model and thinks a lot to match Sol 5.6
The poster argues 763B parameters is GLM 5.3 territory rather than flash-sized, per-task speed is unremarkable, and results match 5.6 Sol and Grok 4.6 only by thinking heavily.
Elshayib
@elshayib_Archived 11 days agoOriginally posted 2026.10.18
Elshayib @elshayib_Originally posted 2026.10.18
DeepSeek v4.1 flash is not FLASH at all The size isn't flash at 763B that's GLM 5.3 size The speed per task also isn't (Not tok/s) Now to be fair it's good and yes same results I've been getting previously with 5.6 sol and Grok 4.6 but for it to achieve that it thinks like a MF.
x.com/elshayib_/status/2111534739903447052Archived 2026.10.18 16:29