Skand Tandon
The open frontier

Four open models. One rack.

In eight weeks the open frontier went from "interesting" to "the thing you deploy." A 1.6T model now fits on an eight-GPU node. A 320B model serves a million tokens of context on four. This page is what it actually costs to run them on Blackwell, and every number links to where it came from.

Will it fit on my box? The drafter that changes the math

The table

sourced, not estimated

Anything marked derived is arithmetic I did myself because the vendor publishes the format but not the byte count. Hover it for the working. Everything else is lifted from a primary source and linked below the card. None of it was measured on my own hardware, which bothers me enough that I wrote down the harness I would run to fix it.

Will it fit?

pick your hardware

The models

what each one is betting on

The drafter

not a model you host

What I take from this

opinion, clearly labelled

Active parameters won

The four models on this page span 284B to 1.6T total, a factor of five and a half. Their active counts span 13B to 49B, a factor of under four, and the two best-value models sit at the bottom of that range. Total parameters decide whether you can host it. Active parameters decide what it costs you per token, forever.

The KV cache is the product

GLM-5.3-Flash's IndexPool gets 4.44x on cache size. DeepSeek V4 reaches 10% of V3.2's cache at 1M context. Neither of those is a footnote. A million-token context is not an attention-math achievement, it is a memory-footprint achievement, and the attention redesign is downstream of that.

Read the licence before the benchmark

GLM-5.3-Flash is MIT. GLM-5.3 is not, it ships under a bespoke licence with the same vendor's name on it. That distinction will cost some team a deployment, and it is one line in a model card that most benchmark roundups do not print.

Speculation stopped being optional

DeepSeek ships MTP and DSpark configs in the recipe. DFlash 2 gets 2.7-4.6x losslessly. When the drafter is in the published serving config rather than a paper appendix, speculative decoding has stopped being a research result and become part of the runtime.