Skip to main content
Model Square is the main place to choose models and understand prices. Billing handles balance and top-ups; Usage handles per-request troubleshooting.

Filter models

Search by provider, tag, and keyword.

Compare prices

Review input, output, cache, per-call, or dynamic pricing.

Copy model names

Copy the exact model name shown in Model Square for the model field.

​
Page information

Model Square usually shows:
  • Model name: the exact value to use in the API model field.
  • Provider: where the model or compatible channel comes from.
  • Tags: short labels for search and rough grouping.
  • Pricing: input, output, cache, per-call, or dynamic pricing.

​
Model names

Use the exact model name shown in Model Square for the model field. Names can be case-sensitive and may include punctuation, slashes, or hyphens.

​
API snippets

API snippets in Model Square are general request templates for OpenAI, Anthropic, or Google-compatible calls. They do not mean that every model has a unique API style. When using a snippet, focus on three things: For your first external call, use the direct HTTP or SDK examples in Quickstart.

​
Tokens and prices

Tokenization depends on the model. Chinese, English, punctuation, code, JSON, and whitespace can produce different token counts. Actual usage is reported in API usage fields and Usage logs. Common pricing types: For token-based models:
Prices are usually displayed per million tokens. Dynamic pricing and task-based pricing should be confirmed from the page pricing notes and final Usage logs.

​
Selection guidance

  • Basic classification, summarization, and formatting: start with lower-cost fast models.
  • Reasoning, coding, and long analysis: choose stronger reasoning or context models.
  • Image understanding: do not rely on tags alone; upload a sample in Conversation to confirm image input support.
  • Document QA: check context length, PDF support, and pricing.
  • Structured data: choose models that support JSON mode or tool calls reliably.

​
Cost controls

  • Keep only necessary conversation history and context.
  • Set max_tokens when output length should be bounded.
  • Use prompt caching when a model supports repeated prefixes.
  • Create separate API keys with quota limits per project.
  • Review high-cost calls in Usage.

​
Next steps