WHAT NEW IN X10.29 THU

AI collections from X · 22 collections · 60,637 posts

deepseek-v4.1POST

DeepSeek V4.1 Flash 1017: 77.0 on DeepSWE, edging Opus 5

Open weights under MIT; scores 77.0 on DeepSWE vs Opus 5's 74.0 at about $0.20 per task (~50x cheaper), but still trails on Terminal-Bench and ProgramBench.

Oguzhan
@0xOgiArchived 11 days agoOriginally posted 2026.10.17
Oguzhan @0xOgiOriginally posted 2026.10.17
Open original on X

77.0 on DeepSWE. DeepSeek-V4.1-Flash-1017 is out today, open weights, MIT. On DeepSeek's chart it edges Opus 5 (74.0) on DeepSWE, about $0.20 a task, roughly 50x cheaper. Opus 5 still leads Terminal-Bench 51.8 to 42.3 and ProgramBench 37.0 to 29.1. Send routine fixes to Flash.

x.com/0xogi/status/2111479884157419832Archived 2026.10.18 16:35