Jensen huang doesn't want you to know you can run a local llm called kimi k3 with 2.8 trillion…

Jensen huang doesn't want you to know you can run a local llm called kimi k3 with 2.8 trillion parameters on the old $500 nvidia card in your closet for $10 instead of $200 opus 5 or $200 chatgpt pro one pip install and every open source llm runs on a gaming gpu: pip install…

starmexAug 206

INSANE. Qwen 3.8 27B is now running locally on an RTX 4060 with just 8GB VRAM.

64,000 token context window using Unsloth's new IQ4_XS quant, only 14.6GB on disk. Prefill hits 150 tokens/sec, decode at 5 tokens/sec via native MTP. Just 25 GPU layers offloaded to stay inside

CyrilXBTAug 2059

Grok 4.6 came out swinging this year — xAI pushed hard on reasoning benchmarks and it's showing.

But Opus 5 still has that edge in depth and nuance for complex tasks. Hard to call honestly. Depends on what you're building. Which one won YOU over in that test? 👀

Gabriel AbiAug 20

Real usage data from someone in my chat

Opus 5 high — 95.2M tokens = 72% of plan Grok 4.6 extra high — 113M tokens = 5.7% of plan Roughly 20x, measured on actual work. If you can't afford the frontier tier, you can still build almost anything. You just can't build something

Apoorv JainAug 20