Models

Models on the platform come from providers such as Qwen (Alibaba), Doubao (ByteDance), Kimi (Moonshot), MiniMax, GLM (Zhipu), DeepSeek, and Kling (Kuaishou), covering six capability categories.

The model list is updated daily: which models exist and what they cost is always governed by the console's Model Plaza in real time. This page doesn't enumerate the list — it explains three things: how models are categorized, how pricing works, and how to read model names.

Model Plaza

Model Categories

CategoryCapabilityTypical use
ChatText understanding and generationChat Q&A, writing, reasoning, code
Vision multimodalImage / video understanding and Q&AScreenshot Q&A, document / chart recognition, video analysis
Image generationText-to-image, image-to-image, image editingPosters, illustrations, image edits
AudioSpeech synthesis, song / BGM generation, speech recognitionVoiceovers, podcasts, music assets
Embedding & rerankText vectorization, retrieval rerankingRAG, semantic search
Video generationText-to-video, image-to-videoShort-video material generation

How to call each category, with code examples, is in API Calls.

Finding Models and Prices in Model Plaza

Model Plaza (console → Model Plaza) is the single authoritative source of model information:

  • Filter: combine filters by provider, group, pricing type, and endpoint type
  • Search: search model names directly by keyword
  • Copy name: click "Copy" on a model card to get the exact name to put in the model parameter
  • Check price: unit prices are shown right on the cards — ¥/million tokens for pay-as-you-go models, ¥/request for per-request models, and each tier's input-length range with its unit price for dynamic-billing models
  • Details: click "Details" on a model card for the full information of that model

Model Plaza detail

Billing Methods

TypeRule
Pay-as-you-goInput / output tokens priced separately; some models also have lower cache prices
Per-requestFixed price per request, independent of token count (mainly image and speech-synthesis models)
DynamicTiered billing by input length: each request is billed at the unit price of the tier it falls into

Charging rules and quota conversion (500,000 quota = ¥1) are in Billing & Top-up.

Groups

When creating an API Key you can pick a group (default / staff); a token can only use models in its group. Filter by group in Model Plaza to see each group's currently available models.

Model Naming Conventions

Name prefixes map to providers:

PrefixProvider
qwen / Qwen / text-embeddingAlibaba Tongyi Qianwen
kimiMoonshot
glmZhipu AI
deepseekDeepSeek
doubao / seedance / seedream / volcengineByteDance Doubao family
MiniMaxMiniMax
klingKuaishou Kling
kling-2 / kling-3 / kling-o1 / kling-motion / kling-v-Kuaishou Kling (unified format)

Common suffixes indicate a model's positioning:

SuffixPositioning
flash / turbo / liteLightweight and fast, lowest price
plusBalanced mainstream tier
max / proFlagship tier — best quality, highest price
previewPreview release; behavior may change
vl / visionVision (image / video understanding) model
editImage-editing model
coder / codeOptimized for code
longExtra-long context
Date (e.g. -2026-03-03)Frozen version of that date
WARNING

Model names are case-sensitive: names differing only in case may be two different models with different prices (e.g. qwen-plus vs. Qwen-plus). Always copy the exact name from Model Plaza before calling.

Multi-Protocol Endpoints

Some chat models simultaneously support OpenAI Chat, OpenAI Responses, Anthropic (/v1/messages), and native Gemini endpoints. Filter by endpoint type in Model Plaza to find the models that currently support multiple protocols; how to connect to each endpoint is in Base URLs & Auth.

Choosing a Model

Match your need to a suffix (meanings in "Model Naming Conventions" above):

NeedLook for
Cheap, high-volume everyday chatflash / turbo / lite
Balanced and good enoughplus
Complex reasoning, best qualitymax / pro
Writing codecoder / code
Long documentslong
Image / video Q&Avl / vision
Text-to-image / image editingimage / seedream (with edit for editing)
Making videosseedance / video
RAG retrievalembedding (vectorize) + rerank (rerank)
TIP

Not sure which to pick? Try them side by side in the console Playground, or start with the cheapest flash series and step up if the quality isn't enough.