Strata, a free desktop app for running AI models locally, has been promoted with claims that the Qwen3.8 Flash Next model can run at more than 20 tokens per second on a graphics card with 8GB of memory.
Local AI · Strata desktop app
Strata claims Qwen3.8 Flash Next runs above 20 tokens per second on an 8GB graphics card
A free app for running AI models on hardware users already own. The headline speed figure has not been independently verified, and key test details were not disclosed.
20+
tokens per second, as claimed by the developers
8GB
graphics memory on the card used in the claim
1-click
installation, according to the developers
Claimed speed on an 8GB card
Each block = 1 token per second. Outlined blocks mark the part of the claim beyond 20 that is not specified.
The claim is a floor of 20 tokens per second for Qwen3.8 Flash Next on an 8GB card.
That speed would be usable for interactive chat and writing on consumer hardware, if it holds in independent testing.
What the announcement states
Latest update makes models faster across all hardware
Largest gains are on lower-cost GPUs
One-click installation
Runs open-weight models locally, without cloud services
What remains unknown
Which GPU model was tested
Quantization level of the model
Context length used
Full list of supported platforms (cut off)
Independent verification of the speed claim
Bottom line
Treat the speed claim as the developers' own assessment until independent tests confirm it.
Quantization, context length and GPU model each significantly change speed and memory use.
Results on other machines may differ.
If the gains hold, mid-range and older graphics cards could run capable local AI models.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...