Playground

Path: Console → Playground /console/playground

What it does

Call any model from the browser, no code. Useful for:

• Iterating on prompts
• Comparing models side by side (GPT-4o vs Claude vs DeepSeek)
• Verifying parameter effects (temperature / top_p)
• Generating ready-to-paste request bodies

Layout

┌─────────────┬─────────────────────────────┐
│  parameters │   conversation               │
│             │                              │
│  Model      │  [user]   ...                │
│  Group      │  [assist] ...                │
│  Temp       │  [user]   ...                │
│  Top P      │                              │
│  Max Tokens │                              │
│  Seed       │                              │
│  Tools      │                              │
│             │  [input ........]  [send]   │
├─────────────┼─────────────────────────────┤
│  Body       │  Code view                   │
│  (JSON)     │  curl / Python / TS         │
└─────────────┴─────────────────────────────┘

Capabilities

• Streaming toggle — see output token-by-token
• Reasoning display — for deepseek-reasoner / qwen-qwq / glm-4-thinking and other reasoning models, shows reasoning blocks
• Direct request body editing — power users edit JSON directly
• One-click code snippet — current request as curl / Python / TypeScript
• Save config — store frequently-used prompt + parameter combos
• Multi-turn — keep history and follow up

Logs integration

Each send creates a logs entry. From Playground you can jump straight to the row to see trace_id, upstream signature, and full usage.

FAQ

Q: Is Playground charged? A: Yes. Same path, same billing as code calls.

Q: Can I upload images / files? A: Images yes (for multimodal models like qwen-vl-max, glm-4v, doubao-vision). File types depend on the chosen model.