One thing that might not be obvious about about DSV4 is how much innovation the Deepseek team implemented in its architecture. When llama.cpp fully supports its lightning indexer, the full 1M context will only require about 6G of RAM. So even though they are similar in size, I believe Deepseek will be much more efficient in that regard.
> I wonder if Hy3 can compete there
Highly depends on how well Hy3 is resilient to quantization. DSV4 is useful even at 2-bit quants.
Its also only 13B active, so your decode speed would be nearly 2x that of Qwen3.6-27B. So there are other latent benefits as well.
https://huggingface.co/collections/z-lab/dflash
I'm running the qwen3.6-27B + dflash on a spark and tgen is way up, but keep the number low, acceptance rate is terrible beyond half a dozen and it requires more memory
For 'general intelligence', DS4 Flash seems to be a noticeable step up still.
And for MoEs, very small amounts of loss can mean you're flipped to entirely different experts (this is also a problem more broadly with numerical stability issues too).
Whereas I can run DSv4 Flash on a pair of DGX Sparks and have enough memory left over for 3M tokens of KV cache, with Hy3 (quantized to FP4), there is only room for ~130K tokens of KV cache.
It's exciting that the open models continue to get better and more efficient across the board!
Edit: fixed, got bad info