Here's the Deepseek 4.1 1017 EXL3 Pagoda shootout with Sushi 1.75 STQ, 2 of turboderp_ 's EXL3s (1.68 and 2.25), and the 2.81 I did.
I personally think its a tossup between Sushi and EXL3 2.25 despite 2.25 preserving the KLD difference. What's also interesting is that 1.75
Mike Adams is releasing a free open-source knowledge base built with local DeepSeek that runs without a GPU.
#LocalAI #HumanControl #ExitAndBuild #DeepSeek #Sovereignty Stay informed on current events, visit
Why are people saying Tencent Hy4 is really good at 3D?
This is Hy4 high vs DeepSeek V4.1 Flash 1017 max making a self contianed HTML using threejs. - Hy4 high: ~55 min, $1.38, 18.89M tokens incl. cache - DeepSeek V4.1 Flash 1017 max: ~3h 13m, $1.22, 143.88M tokens incl.
By the way, with Opus5.5 and Deepseek v4.1Flash (using dsh)
I tried comparing what I created using Three.js As for Opus, it goes without saying that I want to say that it's not very considerate, but as for Deepseek, although it's extremely cheap, I feel that it has reached a certain level. It's just a matter of API conversion.
You have reached the end of the archive
All of deepseek