DeepSeek
DeepSeek V4 and Qwen3.6-27B, both out last week, put pressure on Western models – especially for pricing.
The "stealth" 1-trillion-param model was mistaken for DeepSeek V4. Turns out there was a link.
Are pixels are better inputs to LLMs than text? A novel new model suggests the answer is yes.
"Significant efficiency improvements" claimed in long-context, alongside costs on par with Gemini Flash and gpt-5-nano.
Chinese giants reining back on GPUs ahead of next reasoning model release
"What they did is HARD. Going past CUDA to really get that level of optimization is not a small accomplishment..."
Two "unusual, open ports (8123 & 9000) that led to the exposed database" were not hard to find.
