Spent the last few hours testing Qwen3.8-Flash-Next and GLM-5.3-FLASH locally.
I'm convinced at this point: you can now run near-frontier tier AI on just 1-2 DGX Sparks or a single Mac Studio.
They feel way better to use than Opus-5 after it got heavily nerfed, easily matching pre-nerf Opus-4.8 quality.
If they don't drop Astra and next-gen Fable soon, they're going to bleed market share fast.
Van @jun_song (Jun Song) op X.