Ever spent 45 minutes copy-pasting the same prompt into four different AI models trying to figure out which one actually answers your question well? That's what PromptGrid fixes.
Paste your prompt once. Hit Compare. All 4 NVIDIA NIM models answer at the same time. You see who was fastest, who wrote the most, and read every response side-by-side.
-
Choosing a model for your app — You're building a customer support bot. Drop in your hardest edge-case prompt and see which model handles it cleanly before you commit to one.
-
Debugging a bad response — Your current model is giving weird answers to one specific question. Fire it at 4 models to see if it's your prompt or the model.
-
Quality vs. speed tradeoff — DeepSeek gives a thorough answer in 2 seconds. GLM gives a good-enough answer in 0.8s. PromptGrid shows you both so you can make the call.
-
Writing the right system prompt — Toggle the system prompt field, try different instructions, and see instantly how each model reacts differently.
- You type (or paste) your prompt in the box
- Optionally add a system prompt (click "System Prompt" to expand)
- Adjust max tokens and temperature if needed
- Hit Compare (or Cmd+Enter)
- All 4 models are called in parallel — total wait time = slowest model, not sum of all
- Cards appear with each model's full response, latency in seconds, and token counts
- The fastest model gets a ⚡ badge. The most detailed gets 📝.
- Copy any response with one click, or Download JSON to save everything
- DeepSeek V4 Pro — strong reasoning, good at multi-step problems
- GLM 5.2 — fast, efficient, great for quick answers
- Mistral Medium 3.5 — balanced, strong at instruction following
- MiniMax M3 — multimodal-capable, long context
pip install -r requirements.txt
uvicorn main:app --port 8008
# open http://localhost:8008Requires NVIDIA NIM API key in ~/keys/api_keys.py with NVIDIA_KEYS list.
FastAPI + httpx async + vanilla JS single-page UI. No build step, no npm, no framework.