Kasoku

Browser GPU test20 seconds

See what AI your GPU can actually run.

Kasoku tells you which local AI models your graphics card can run. No account.

When a test finishes, the card model, score, tier, browser and OS go to the public leaderboard. Nothing that identifies you. Privacy

Adapter

Looking for a GPU

Detecting

Fig. 1 Thermal view of your GPU

Five tiers, from potato to beast.

A card lands on the lower of its speed and its memory. Each tier adds models you can run today. How tiers work

  1. Potato

    Any card

    Tiny models and simple tools. Still more than nothing.

    • Gemma 3 1B
    • Moondream 2
  2. Starter

    250+ and 3 GB+

    Small chat models, speech to text and background removal.

    • Llama 3.2 3B
    • Phi-4-mini
  3. Solid

    900+ and 6 GB+

    Everyday 3-4B chat models and quick image tools.

    • Gemma 4 E4B
    • Llama 3.1 8B
  4. Strong

    2,400+ and 8 GB+

    7-8B chat models at reading speed and local image generation.

    • FLUX.1 schnell
  5. Beast

    6,000+ and 16 GB+

    Larger models, SDXL-class images and long context.

    • gpt-oss 20B