X · @teortaxesTex
· X / Twitter
to be clear: DeepSeek-V3.2 scored 4.0% at 3 times the cost of V4-Flash's 61.4%. This is 9 months of progress (give or take; V3.2 Speciale not tested, …
to be clear: DeepSeek-V3.2 scored 4.0% at 3 times the cost of V4-Flash's 61.4%. This is 9 months of progress (give or take; V3.2 Speciale not tested, but V4-Pro Official not yet available). I can see them reaching Fable scores (likely at a lower cost) by the end of this year.ARC Prize: DeepSeek V4 Flash from @deepseek_