Building AI applications means testing dozens of models, comparing outputs, tuning parameters, and understanding exactly what happens under the hood. The ANT Playground is built for exactly this. It's not a demo — it's a production-grade testing lab that gives you direct access to thousands of models across 40+ providers, all through a single interface.
One Interface, Thousands of Models
The Playground connects to every provider ANT supports: OpenAI, Anthropic, Google, DeepSeek, Mistral, Meta Llama, Qwen, Cohere, and dozens more. You can switch between models instantly — from GPT-4o to Claude to Gemini to Llama 4 Maverick — and see how each one handles the same prompt. No API keys required for the free tier. Log in to unlock the full catalog.
Virtual models like ant:auto, ant:fast, ant:best, and ant:reasoning let ANT's routing engine pick the best model for each request automatically. Or select a specific model from any provider to test it directly.
Fusion Mode: Side-by-Side Model Comparison
Fusion mode is where the Playground becomes a real evaluation tool. Select 2-8 models and send the same prompt to all of them simultaneously. Each response streams in its own column with real-time metrics: time to first token (TTFT), total duration, and token count. You can instantly see which model gives the best answer, which is fastest, and how outputs differ.
This is invaluable for prompt engineering. Test a prompt against GPT-4o, Claude Sonnet, Gemini 2.5 Flash, and DeepSeek V3 at the same time. Find the model that handles your use case best, then configure your ANT routing to prefer it.
API Inspector: See What Happens Under the Hood
Every request you make in the Playground is fully visible in the API Inspector panel. Click the Inspector tab on the right to see the exact JSON request body, response headers, HTTP status code, timing metrics, and the model that actually served your request. You can copy any request as a curl command with one click.
The Compose tab takes this further — write raw API calls in JSON, fire them directly, and see the results stream into the chat on the left. This bidirectional flow means you can prototype in the chat, inspect the API call, modify it in the Compose tab, and test variations without leaving the Playground.
Request Topology: Watch the Routing Pipeline
The Topology tab shows ANT's classification and routing pipeline in real-time. For every request, you can see each stage: received, preprocessed, classified, optimized, routed, streaming, completed. Each stage shows timing deltas and metadata — what ANT classified the request as, which models were considered, why a specific model was chosen, and how long each stage took.
This is the kind of visibility you don't get from any other AI gateway. You see exactly why ANT chose GPT-4o over Claude for a particular request, or why it routed to Groq for a simple task.
Vision, OCR, and Multimodal Testing
Switch to Vision mode and drag-and-drop images directly into the chat. ANT routes to vision-capable models automatically. Upload screenshots, diagrams, documents, or photos and test how different models interpret them. OCR mode is optimized for document extraction — upload a scanned document and compare extraction quality across models.
Images can be pasted from clipboard, dragged in, or uploaded via the file picker. They're sent as base64 data URIs, so no external hosting needed. The Playground handles the OpenAI-compatible image_url content format automatically.
Nests: Custom Model Groups
Nests are one of ANT's unique features. Create a custom group of models — for example, a 'coding' nest with DeepSeek V3, Claude Sonnet, and GPT-4o — and route to that nest in the Playground. ANT will pick the best model from your nest for each request. This lets you build custom routing strategies without writing code.
In the Playground, select any nest from the model dropdown. Nests are created and managed in the dashboard. Each nest can have its own model mix, and ANT's routing engine evaluates all models in the nest for every request.
ANT Search: Grounded Responses
Select ant:search as your model and ANT will augment the response with real-time web search results via the Sonar API. Responses include inline citations so you can verify claims. This is useful for testing how well models integrate external knowledge with their own reasoning.
Building Applications with the Playground
The Playground isn't just for testing — it's a development tool. Use the Compose tab to prototype the exact API calls your application will make. Test different models, temperatures, and system prompts. Copy the working request as curl and integrate it into your codebase. The API is fully OpenAI-compatible, so any SDK or tool that speaks the OpenAI format works with ANT.
# Copy from Playground's Inspector tab and use directly
curl https://antbase.ai/v1/chat/completions \
-H 'Authorization: Bearer ant_...' \
-H 'Content-Type: application/json' \
-d '{
"model": "ant:auto",
"messages": [{"role": "user", "content": "Explain quantum computing"}],
"temperature": 0.7,
"stream": true
}'Access for Everyone
The Playground is free to use without an account — you get access to ant:free which routes to the best available free models from providers like Groq, NVIDIA, Gemini, Cerebras, and more. Sign up to unlock the full model catalog, Fusion mode, Nests, and your own API keys. Paid credits give you access to premium models from OpenAI, Anthropic, and others.
- •Anonymous: ant:free with rate limiting — no signup needed
- •Free account: All virtual models, up to 2 models in Fusion, full catalog browsing
- •Credits: All models, 8 models in Fusion, Nests, API keys, ant:search
Try It Now
The ANT Playground is live at antbase.ai/playground. Open it, pick a model, and start chatting. Toggle the API Inspector to see the raw calls. Switch to Fusion mode to compare models. No setup required.
