A free DeepSeek model hit 73 percent on simulated hacking tasks.
OpenAI's gated cyber model scored 81 percent on the same tests. - Vals says it cost two to three dollars to run DeepSeek, versus seventeen to forty-six for GPT 6 Sol. - A security firm says a China-based group is
A 5.6GB uncensored AI model that refuses almost nothing?!
🤯 Meet DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED a hybrid model that reportedly combines DeepSeek-V4-Pro distillation with Qwen3.5 9B. What makes it interesting: Only 5.6GB in Q4_K_M GGUF format Just 3% refusal rate in the
Qwen 4 vs Deepseek Flash
Neck and neck I know this isn't really a challenging scene or anything I've just had a recent obsession with this look. Even on tokens they're pretty close.
“ZZZ Mini-Theater | These little guys are just too adorable—I can’t even handle it!
⸜(｡˃ ᵕ ˂ )⸝♡”
You have reached the end of the archive
All of deepseek