Apple unveiled the mac mini with M6
Their first 2NM chip starting at $899, and the base model runs qwen 3.6 27B and deepseek R1 32B at 13.5X faster LLM processing than M1 while the mac studio M5 ultra handles frontier-scale models entirely in local memory. Apple's Mac business
It's been like this today.
I know it is no DeepSeek speed but still is faster than what I'm used to
Qwen 3.8 27B
Deepseek Harness Time to complete: 13 1/2 minutes Average tokens per second: 369 RTX5090
OpenAI officially unveiled its first self-developed inference chip Jalapeño.
In specific inference tests such as GPT‑OSS 120B, DeepSeek R1 and Kimi K2.5, Jalapeño’s peak throughput per watt increased by 1.5–1.9 times, and end-to-end latency was reduced by 1.7–3.6 times, with some indicators exceeding NVIDIA GB300.
You have reached the end of the archive
All of deepseek