Runs in your browser
WebGPURun a small LLM on this device
Download a model and chat on this device. The downloaded model can be reused next time.
- Model
- Qwen 3.5 0.8B Q4
- GPU memory
- About 1.6 GB
This is a small model; answers may be incorrect or awkward.
The model downloads only after you press Start.
Local AI generation speed
How does my computer rank?
- Chipset / CPU
- Detected when measuring
- RAM
- Capacity not available
- Logical processors
- Closed
- GPU reported by the browser
- Checking
Enter or edit chipset / RAM
On Mac, check Apple menu → About This Mac. On Windows, check Settings → System → About.
Browser RAM may be rounded or capped. A manually entered installed capacity takes precedence.
You can enter missing specifications. They stay in this browser unless you consent to submit a record.
Measurement and ranking method
Qwen 3.5 0.8B Q4, 4-bit · WebLLM 0.2.85. After 1 warm-up, the benchmark measures 128-token decoding 3 times and takes the median. Every language uses the same fixed Korean prompt. First-token wait and model download time are excluded. The reference rank is a comparison scale, not actual user statistics.
Participant rank uses the last 90days of visitor-submitted records, not independently verified measurements. Only the latest record per browser counts. Browser, power and thermal conditions affect results; they do not predict native-app or larger-model speed.