the biggest open AI model ever drops this week. you can download all 1.4 terabytes — you just can't run it

Kimi K3's open weights land July 27. It's genuinely great and genuinely free. It also needs about 1.4TB of memory to load, which means the download…

Aliteq
Lena Fischer · AI & Local Compute Editor

The short answer

Kimi K3 is a 2.8-trillion-parameter open-weight Mixture-of-Experts model from Moonshot AI, with full weights releasing July 27, 2026. Quantized to MXFP4 the weights are ~1.4TB (≈5.6TB at FP16), so…

Open weights July 27, 2026 — the largest open model ever, 1 on the Frontend Code Arena (1,679 points).

~1.4TB at MXFP4, ~5.6TB at FP16. No single consumer GPU, and not even a 512GB Mac Studio, can hold the weights.

The MoE trap: only ~50B of 2.8T parameters are active per token (16 of 896 experts), so compute is manageable — but every parameter must still sit in memory. MoE saves compute, not capacity.

What you'd need: roughly a full 8×B200 (1,440GB) datacenter node, or an equivalent multi-node setup, just to load it.

What you can run instead: capable open models like Qwen3-32B on a single 24GB card get most people most of the way.

Aliteq

Read the full story

the biggest open AI model ever drops this week. you can download all 1.4 terabytes — you just can't run it

Read the full story on Aliteq