Qwen

qwen after cooking 😂

https://x.com/alibaba_qwen/status/1872155286796177852

I don’t know how closed models without specialized inference silicon are going to compete with this

🚀 Introducing DeepSeek-V3!

Biggest leap forward yet:
⚡ 60 tokens/second (3x faster than V2!)
💪 Enhanced capabilities
🛠 API compatibility intact
🌍 Fully open-source models &amp; papers

🐋 1/n https://t.co/p1dV9gJ2Sd

DeepSeek

Curious | building robots | https://anurajrp.io/about/

I don’t know how closed models without specialized inference silicon are going to compete with this

https://x.com/deepseek_ai/status/1872242657348710721

I was watching this Dwarkesh podcast with semianalysis and asianometry guys. Dylan mentions somewhere in between that the data center architecture starts differing based on the silicon used. So I would expect cross pollination and rapid iterative improvements on open models servers than closed ones.

But this was a speculation by looking from outside.

Blogger at ribbonfarm.com, running summerofprotocols.com, host of /gloom ☀️

I was watching this Dwarkesh podcast with semianalysis and asianometry guys. Dylan mentions somewhere in between that the data center architecture starts differing based on the silicon used. So I would expect cross pollination and rapid iterative improvements on open models servers than closed ones.

But this was a speculation by looking from outside. 


https://youtu.be/J7yvSkIOZLw?si=WDj6W_j9Tvyg9QLN

I don’t follow your logic. Model openness and performance are different vectors. 

Also tps doesn’t tell you anything about output quality.