I had the local Qwen3.8-27B on my DGX Spark make a landing page introducing itself
As I scroll down I wanted the site the have smooth transitions and interlinked elements. I've done this general task with many other models, but this was pretty impressive for a smaller local
Fable 5 vs DeepSeek V4 Pro vs Grok 4.6: the number nobody's talking about is 276.
Not the intelligence score. The cache read gap. Agents re-read your instructions on EVERY step. Hundreds of times per task. DeepSeek's cache reads cost ~276x less than Fable 5's. With a 92% hit
April: V4-Pro launches at $3.48 per million output
May: DeepSeek cuts it 75% and calls the discount permanent today: $3.96 at peak the price is now higher than it was before the discount 😭
Local showdown: Qwen vs Deepseek
Fireworks one-shot: "build a fireworks show over a city, single HTML file, no libraries." two locals on DGX Sparks, temp 0.6, one attempt, no edits. DeepSeek-V4-Flash (2x GB10): 1m20s, 5,214 tokens, Qwen3.8-27B NVFP4 (1x GB10, xhigh): thought
You have reached the end of the archive
All of deepseek