Skip to content
Main Site News Console

DeepSeek's New AI Is a Performance Beast

· Two Minute Papers Translated
视频论文

DeepSeek’s New AI Is an Absolute Beast

Image 7: Two Minute Papers Two Minute Papers

102,224 views • Oct 28, 2026

❤️ Learn more about Lambda and sign up for their GPU Cloud here: https://lambda.ai/papers
📝 DeepSeek v4.1 Flash 1017 is available here: …

158 Comments

@Zex-4729

16 hours ago

One of my favorite channels on modern YouTube!

@Wanjalawassim

18 hours ago

0:20Image 13: 😂Image 14: 😂Image 15: 😂Image 16: 😂 The iPhone issue

@differntname2807

16 hours ago

0:21 What a combo! LMAO Image 18: 😂Image 19: 😂

@entalon

18 hours ago

What a time to be alive and prepare for unemployment!

@thelazy0ne

18 hours ago

That brutal burn on the iPhone Image 22: 😂.

@SeanLumly

17 hours ago

All things considered, DeepSeek 4.1 Flash is probably the best LLM out there right now.

@Senumunu

15 hours ago

The first AI capable of producing high-quality animations in Blender will be an absolute game changer.

@RoboArtCo

18 hours ago

What an awesome time to be alive! So early to this.

@vantasoft

13 hours ago

Artificial Analysis’s data also highlights the issue of per-task cost. For Kimi K3.1 Pro and GPT-6.1 Sol (Max), the published pricing for both is $2 per million input tokens and $10 per million output tokens. However, when running the Intelligence Index benchmark, Kimi generated 110 million output tokens, whereas Sol produced only 64 million tokens; the average cost per task came out to $3.68 for Kimi versus just $1.27 for Sol. This means that despite having what appear to be identical rate cards, the actual costs differ by nearly 3x.

@rachel_rexxx

15 hours ago

1:37 Even if it consumes 3x the tokens, if the unit price is 1/15th, the total cost is still only 1/5th of Claude Opus 5.5.

@PatrikZero

18 hours ago (edited)