Test every AI model.
On your machine.

No catch. The app is free under the MIT License. The only cost you'll ever pay is to the AI providers themselves — you're using your own API keys, so you're billed directly by OpenAI, Anthropic, Runware, etc. at their standard rates. If you only use Ollama (local inference), you pay nothing at all.

Entirely on your machine, in a local SQLite database. They are never sent to any server other than the provider you're calling. The app has no backend. We track session and usage so it help me to make the software better. You can inspect the database file yourself at any time.

No. Add keys only for the providers you actually use. Ollama requires no key at all since it runs models locally. You can start with a single provider and add more whenever you need them — the dashboard shows connection status for all 11 at a glance.

Ollama itself handles the hardware side — AI Playground just connects to it. For comfortable local inference, 8 GB of VRAM handles most 7B–13B models well (e.g. Llama 3, Mistral, Gemma). Smaller models like Phi-3 run fine on CPU-only machines. If you have Ollama already installed and running, AI Playground will detect it automatically.

Open the Compare tab, add up to 3 model columns (any provider, any model), type one prompt, and hit Run All. The same prompt fires simultaneously to each model. Results appear in parallel columns as they stream in — you can see which model responds first, compare quality, and check cost side by side. Works the same way for image generation.

Windows, Linux and macOS. The app is built on Electron + Next.js, so cross-platform support is straightforward.

Yes. The History page has an Export button that downloads your run history as a CSV — provider, model, prompt, response, tokens, cost, and timestamp. Because everything is stored in a standard SQLite file, you can also open it directly with any SQLite browser if you need deeper access.

Cost is calculated from the actual token counts returned by each provider's API response, multiplied by that model's current price from the live pricing catalog. It's a close estimate — within a rounding error of your actual provider invoice. For image generation, cost is taken directly from Runware's response payload. Ollama always shows $0.00.

The app is fully translated into 6 languages: English, Spanish (Español), French (Français), German (Deutsch), Brazilian Portuguese (Português), and Simplified Chinese (中文). Every label, button, tooltip, placeholder, and error message is translated. Pick your preferred language in Settings → Appearance, or the app will auto-detect your OS language on first launch.