@thatlev: DeepSeek released V4.1-Flash-1017 with a reported DeepSWE v1.1 score of 77.0 against Claude Opus 5’s 74.0. The new model is available on its website, app and API. These are DeepSeek’s own results, and Claude leads several other displayed benchmarks. The DeepSWE victory speech belongs to DeepSeek #DeepSeek #Claude #DeepSWE #AIBenchmarks #AINews
deepseek v4.1 flash is trash. very stupid, thinks always to much and spends many tokens af
2026-10-18 18:47:21
0
Bartekowicz :
gemini 3.8 flash beats everyone but realisticly its ass
deepswe worst test tbh
2026-10-18 16:41:48
0
nu metal pigeon :
What nice ai can run on a 4GB GTX 1650? either way I'm cooked on VRAM, I know.
2026-10-18 12:06:26
0
Bp🔨 :
Why are the people bad asf g
2026-10-18 11:53:45
1
Maximized :
opus 5.5 xlow deepseek 4.1 max
2026-10-18 16:26:44
0
Karelian :
i want them
2026-10-18 11:11:56
1
Novaws :
early mommy
2026-10-18 11:18:50
1
real.senku :
make them kiss now
2026-10-18 16:33:56
0
mustsnerd :
Hehe
2026-10-18 15:44:32
0
cy :
I currently configured my hermes agent with haiku 5.5 as main and mimo 2.6 flash for delegation. But I'm still trying to figure out what actually works well for me in practice. Results are meh but haiku seems better than gpt for me.
2026-10-18 14:51:52
0
нигодник :
I would do deep seek in DeepSeek
2026-10-18 11:31:41
0
Venti :
gemini 4 argon ?
2026-10-18 13:16:54
0
Uni :
cap
2026-10-18 13:35:29
0
True MTAC Enjoyer :
And smarter [Слезы радости]
2026-10-18 11:22:16
0
Roman :
tell me the prompt
2026-10-18 13:05:07
0
E. Gierek :
goons intensively
2026-10-18 13:01:54
1
DuXBITH :
how deep can I go in DeepSeek?
2026-10-18 13:47:27
2
wooden horse :
dsv4.1f turns into a schizophrenic 5 year old when you give it literally any somewhat long horizon task, blows cheeks
2026-10-18 11:29:08
2
To see more videos from user @thatlev, please go to the Tikwm
homepage.