gpt-6POST
DeepSeek V4.1 Flash scores 74.2% on DeepSWE, beating GPT-6 Astra and 5.6 Sol
A leak claims the 552B DeepSeek V4.1 Flash reaches 74.2% on DeepSWE at a very low price, ahead of GPT-6 Astra, Opus 5 and 5.6 Sol.
Haider.
@haider1Archived 2026.09.16Originally posted 2026.09.10
Haider. @haider1Originally posted 2026.09.10
how tf does deepseek v4.1-flash, a 552b model costing around $0.15/m input + $0.60/m output score 74.2% on DeepSWE beating gpt-6 astra, opus 5, 5.6 sol, fable 5 and even fable 5.1 we're talking about a model competing with trillion-scale frontier systems for a fraction of the price
This post quotes another post: x.com/deepseek_ai/status/2097930608790167907
x.com/haider1/status/2098006898717163576Archived 2026.09.16 19:34