Qwen3.8-Flash-Next

The one model Strata is built around. It's a mixture of experts: 30B parameters in total, but only 3B work on each word, so it runs fast on a home PC.

30B total · 3B active Up to 1M context Text · code · tools

Loading {{ switchName }}

Running {{ runName }} {{ runQuant }}

{{ runSpeed }} tok/s

Tune engine

Not loaded Nothing is using your graphics card right now.

Choose a version

Same model, packed at different sizes. Smaller versions are faster and use less memory; bigger ones are a little more precise.

Fast on your PC Works, but slower Too big for your PC

← Smaller and fasterMore precise →

{{ sel.id }} {{ sel.chip }}

Quality {{ sel.quality }}% of original

Writing speed {{ sel.speed }} tok/s

Download size {{ sel.size }} GB {{ sel.diskNote }}

Memory needed {{ sel.need }} GB

Technical details

              bits / weight{{ sel.bits }}
              perplexity vs BF16{{ sel.ppl }}
              experts in VRAM{{ sel.expertsVram }}
              file{{ sel.file }}
            

This version is running All chats use it. Tune speed and memory on the Engine page.

Downloaded and ready {{ act.switchNote }}

Downloading{{ act.dlPct }}%

Not downloaded yet {{ act.availNote }}

Too big for this PC {{ act.bigNote }}

On this PC

{{ row.name }} {{ row.id }} {{ row.meta }}

Running

{{ row.dlPct }}%

Model

Qwen3.8-Flash-Next

Choose a version

{{ sel.name }}

On this PC