A2Agent gives you access to 20+ models across 5 major Chinese AI platforms — DeepSeek, GLM (智谱), Kimi (Moonshot), MiniMax, and Qwen (通义千问) — through a single OpenAI-compatible API. Every model below is ready to use with your A2Agent API key; no separate platform accounts or credentials required.
Models by Provider
DeepSeek
GLM
Kimi
MiniMax
Qwen
DeepSeek models excel at both general-purpose chat and multi-step reasoning tasks, making them a strong default choice for coding assistants, analysis pipelines, and interactive applications.Model notes
deepseek-v4-flash — Fast, cost-efficient variant of the V4 series. Best for high-throughput workloads where response speed matters.deepseek-v4-pro — Full-capability V4 model with stronger reasoning performance. Recommended for complex, multi-step tasks.
GLM (智谱 AI) models offer long-context windows up to 205K tokens and span general chat, agentic workflows, and advanced coding. The GLM-5 series introduces native agent-mode support for tool-calling pipelines.Model notes
glm-4.6 / glm-4.7 — Fourth-generation GLM models with 205K context; reliable for document-heavy chat and summarization.glm-5 — Fifth-generation flagship model with agent capabilities for tool-calling and multi-step orchestration.glm-5-turbo — Faster, slightly pricier GLM-5 variant optimized for interactive agentic use cases.glm-5.1 — Tuned for code generation and technical tasks alongside general chat.glm-5.2 — Lightweight GLM-5 variant for cost-sensitive chat workloads.
Kimi (Moonshot AI) models stand out with their 256K context window and native vision support, making them ideal for document analysis, image understanding, and long-form conversation.Model notes
kimi-k2.5 — Cost-effective Kimi model with full 256K context and vision input; great for PDF and image-grounded Q&A.kimi-k2.6 — Enhanced K2 model with improved instruction-following and multimodal understanding.kimi-k2.7-code — Code-specialized Kimi variant optimized for generation, review, and debugging tasks.
MiniMax M2 models deliver a balanced combination of long context (200K), coding ability, and competitive pricing, making them a practical choice for production deployments that need scalability without high costs.Model notes
minimax-m2.1 — Entry-level M2 model; best for bulk processing and straightforward coding tasks.minimax-m2.5 — Mid-tier M2 with improved coding quality at the same price point as M2.1.minimax-m2.7 — Latest MiniMax model focused on general chat with 200K context support.
Qwen (通义千问) models from Alibaba Cloud span a wide capability range — from lean MoE architectures to frontier-class reasoning — giving you fine-grained control over the cost-performance tradeoff.Model notes
qwen3.5-plus — Qwen 3.5 flagship with strong general reasoning and instruction-following.qwen3.6-35b-a3b — 35B dense model with a 3B active parameter MoE design; highly efficient for its capability tier.qwen3.6-plus — Balanced Qwen 3.6 variant offering low cost with solid output quality.qwen3.7-max — Top-of-line Qwen 3.7 model for the most demanding reasoning and generation tasks.qwen3.7-plus — Performance-efficient Qwen 3.7 variant; great for high-volume workloads.
Model Type Glossary
Understanding each capability type helps you choose the right model for your task.
- chat — General-purpose conversational inference. Covers Q&A, summarization, translation, creative writing, and instruction-following.
- reasoning — Extended chain-of-thought capability for complex logic, math, and multi-step problem solving. These models think before they respond.
- coding — Fine-tuned for source code generation, debugging, code review, and technical documentation.
- agent — Supports tool-calling and function-calling protocols for building autonomous agents and multi-step orchestration pipelines.
- vision — Accepts image inputs alongside text prompts. Use for document parsing, screenshot analysis, and image-grounded Q&A.
- long-context — Models with context windows of 128K tokens or more. Suitable for processing entire codebases, books, or lengthy conversation histories in a single request.
Model IDs are case-sensitive. Use them exactly as shown in the tables above when setting the model parameter in your API requests. For example, use deepseek-v4-pro, not DeepSeek-V4-Pro or deepseek-v4-Pro.
Log in to your A2Agent account to unlock exclusive channel pricing — lower per-token rates on select models that aren’t visible to unauthenticated users. View your pricing → 
