logo
Development
OverviewDevSpaceWorkSpaceAPI ReferenceBest PracticeChangelog
Search
English
WorkSpace / Appendix / Supported Models
Supported Models
Last updated：4 months ago

Supported Models

Overview

The workspace supports multiple LLM providers through a provider-agnostic adapter. The following lists all preset models and their capabilities.

Model Capability Matrix

Model Provider Context Window Vision Document Audio Tool Use Recommended Scenario
Claude Sonnet 4 Anthropic 200K ✅ ✅ — ✅ Balanced first choice, ideal for daily use and coding
Claude Opus 4 Anthropic 200K ✅ ✅ — ✅ Strongest reasoning, complex analysis tasks
Claude 4 Anthropic 200K ✅ ✅ — ✅ All-rounder, great value for money
Claude 3.5 Sonnet Anthropic 200K ✅ ✅ — ✅ Stable and reliable, good compatibility
Claude 3.5 Haiku Anthropic 200K ✅ — — ✅ Fastest response, simple tasks
GPT-4o OpenAI 128K ✅ — ✅ ✅ Audio support, rich multimodal
GPT-4o-mini OpenAI 128K ✅ — ✅ ✅ Lowest-cost daily conversation
GPT-4-turbo OpenAI 128K ✅ — — ✅ Classic general-purpose model
GPT-4.1 OpenAI 1M ✅ — — ✅ Ultra-long context, whole-repo analysis
o3 OpenAI 200K ✅ — — ✅ Deep math/logic reasoning
o4-mini OpenAI 200K ✅ — — ✅ Lightweight reasoning tasks
DeepSeek-V4-Flash DeepSeek 1M — — — ✅ Code generation, extremely low cost
DeepSeek-V4-Pro DeepSeek 1M — — — ✅ Pure reasoning (no tool calls)
Qwen-plus Qwen 128K ✅ — — ✅ Chinese-optimized, low cost
Qwen-max Qwen 128K ✅ — — ✅ Strongest Chinese, complex Chinese tasks

Capability Descriptions

Vision (Image Understanding)

Supports attaching images to messages, and the Agent can understand and analyze image content.

  • Anthropic: Native base64 image content blocks
  • OpenAI: image_url (data URI format)
  • Unsupported models: Image inputs are filtered out or reported as unsupported

Document (Document Understanding)

Supports directly passing in documents such as PDFs, and the Agent can read and analyze document content.

  • Anthropic: Native base64 document content blocks
  • Other providers: Downgraded to a text description or reported as unsupported

Audio (Audio Understanding)

Supports passing in audio content.

  • OpenAI: Native audio content blocks
  • Other providers: Not yet supported

Tool Use

Supports function calling / tool use, and the Agent can invoke tools to perform operations.

API Formats

Provider API Format Description
Anthropic anthropic Anthropic Messages API
OpenAI openai OpenAI Chat Completions / Responses API
DeepSeek openai OpenAI-compatible format
Qwen openai OpenAI-compatible format
Custom openai / anthropic Choose based on the provider

Custom Models

Users can add non-preset models in Model Configuration, which requires manual configuration:

Configuration Item Description
Model Name LLM model identifier
Base URL API endpoint address
API Key Access key
API Format anthropic or openai
Context Window Number of tokens
Capability Flags vision / document / audio / tool_use