← sources

Lowest $/tok/s (or highest tok/s/$)

source: reddit (reddit:LocalLLaMA)published 2026-09-18 14:24ingested 2026-09-19 02:42

retained text

People are in awe of the Qwen3.8-Flash-Next, which pushes me into running towards buying asap (I know APIs are cheaper, but my privacy needs locality, while my wallet needs optimization)

I understand it is an MoE, and probably the future trends will see cheap hardwares being able to achieve a high intelligence thanks more to MoE than to a densed model ?

So there comes my question: which of the hardware config can deliver Qwen3.8-Flash-Next at the lowest $/tok/s ? A cheap system is good for the wallet, but too cheap means too low speed, therefore that unit (Although I know the price will change once people realize this, also after this post)

Next to the common RTX 3090/4090, there are also AMD Mi50, Nvidia v100, Nvidia CMP 170hx, or some other AMDs.

Which paths to go, that also allow upgradability ?

Or I should just not predict the future, just have a system for now (Qwen 3.8), and will need to sell it to buy something else ?

Thanks

document fetched 2026-09-19 00:40 · canonical https://reddit.com/r/LocalLLaMA/comments/1wjrnwx/lowest_toks_or_highest_toks