Models

DeepSeek V4 Pro is the larger V4 route for harder reasoning, coding and long-context analysis workloads.

Qwen3.8 2.4T A95B is the open-weight Qwen3.8 flagship route for million-token reasoning, coding and agent workloads.

Qwen3.7 Flash is the current low-latency Qwen route with a 1M context window, hybrid thinking and a three-tier input price schedule.

Qwen3.7 Max is the top-tier Qwen route for advanced reasoning, code generation and complex tool-using workloads.

Qwen3.7 Plus is the current recommended Qwen route for coding agents, multimodal long-context requests and structured assistant workflows.

DeepSeek Chat is a general-purpose conversational model with strong coding, reasoning and Chinese-English instruction following.

DeepSeek Reasoner is optimized for visible reasoning, complex problem solving and multi-step technical work.

DeepSeek V4 Flash is the low-cost V4 route with 1M context, thinking and non-thinking modes, tool calls and JSON output.

Direct DeepSeek API route for comparing token accounting, latency and supplier charges with the xh.v1api.cc route.

Qwen3.6 Max Preview remains available for existing preview workflows, but Alibaba Cloud has scheduled it for retirement on September 8, 2026.

Qwen3.6 Flash keeps the 1M context and tool support of the 3.6 family at a lower token price.

Qwen Plus is kept as a compatibility route and currently points traffic at Qwen3.6 Plus; new builds should prefer Qwen3.7 Plus.

Qwen3.6 Plus remains a stable Qwen long-context route for agentic coding, multimodal prompts and production assistant workflows.

Kimi K2 is kept as a compatibility route and currently points new traffic at Kimi K2.6.

Kimi K2.6 is Moonshot's latest multimodal agent model for long-horizon coding, visual understanding, deep reasoning and autonomous tool workflows.

Kimi K2.5 is a balanced multimodal model for text, image and video input with thinking and non-thinking modes.

表达严谨，适合撰写正式公文报告与常规工作文档梳理

输出更快，适合批量生成、日常文案和轻量办公任务

推理与表达更充分，适合复杂分析、方案拆解和正式写作

小米 MiMo 全模态模型，适合文本生成、全模态理解、推理、流式输出和工具调用

小米 MiMo Pro 模型，适合长上下文推理、工具调用、结构化输出和联网搜索任务

响应轻快，适合日常对话、快速整理与轻量写作

阿里云百炼官方 Paraformer 实时语音转写模型，按音频秒计费，计划于 2026-10-10 下线

Qwen3.8 Max is the current flagship Qwen route for frontier reasoning, multimodal understanding, coding and complex tool-using workloads.

讯飞星火旗舰非思考模型，适合结构化抽取、逻辑问答和正式办公写作

讯飞星火轻量模型，适合快速对话、草稿生成和低延迟助手交互

讯飞星火专业长上下文模型，适合文档阅读、知识问答和办公摘要

科大讯飞实时语音转写大模型，来自 new_rta 服务，用于 FinClaw 语音输入