🧠机器之心•較早收集於 13m
阿里雲推出最強Coding Plan
💡Access 4 top open-source coding models in one cheap sub; beats single-model limits (global exclusive)
⚡ 30-Second TL;DR
有什麼變化
Coding Plan支持Qwen3.5（3970億參數，激活170億）、GLM-5、MiniMax M2.5、Kimi K2.5無縫切換
為什麼重要
大幅降低高頻編程成本，讓開發者無單一模型限制使用頂尖模型。促進中國開源大模型在全球工具中的採用。
下一步行動
Subscribe to Aliyun Coding Plan Lite via https://www.aliyun.com/benefit/scene/codingplan to test Qwen3.5 on complex coding tasks.
誰應關注：Developers & AI Engineers
關鍵要點
- •Coding Plan支持Qwen3.5（3970億參數，激活170億）、GLM-5、MiniMax M2.5、Kimi K2.5無縫切換
- •Lite每月18,000次請求，首月7.9元；Pro每月90,000次，首月39.9元
- •全球獨家服務，兼容Qwen Code、Claude Code、Cline、OpenClaw
- •平台提供100多款模型API及400多個Agent模板
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 8 個來源。
🔑 增強重點摘要
- •Alibaba Cloud's Coding Plan supports 8 total coding models including both Qwen proprietary models (qwen3.5-plus, qwen3-max-2026-01-23, qwen3-coder-next, qwen3-coder-plus) and third-party models (MiniMax-M2.5, glm-5, glm-4.7, kimi-k2.5), enabling developers to leverage specialized code-generation architectures alongside general-purpose LLMs[3].
- •The Qwen3-Coder-Next model is specifically engineered as an open-weight language model designed for coding agents and local development environments, representing a shift toward edge-deployable coding assistance beyond cloud-only solutions[8].
- •Coding Plan operates exclusively in the Singapore region with global deployment mode, meaning user input and output data may cross borders—a critical compliance consideration for enterprises handling sensitive code[3].
- •The qwen3-max-2026-01-23 flagship model integrates thinking mode with three built-in tools (web search, web extractor, code interpreter) to improve accuracy on complex coding problems through external tool reasoning during inference[5].
🛠️ 技術深入
- Qwen3.5-Plus Pricing Structure: Tiered token-based pricing with input cost of $0.4 per 1K tokens (0–256K range) and $0.5 per 1K tokens (256K–1M range); output costs $2.4 and $3.0 respectively[5].
- Model Context Windows: Qwen3-max-2026-01-23 supports 1,000,000 token context length with 65,536 max output tokens and 81,920 max chain-of-thought tokens[5].
- Batch Processing Optimization: Qwen-plus models offer 50% discount on batch API calls, reducing costs for non-real-time coding workflows[5].
- Token Quota System: Lite plan provides 18,000 monthly requests; Pro plan provides 90,000 monthly requests, with new user 80% first-month discount applied to both tiers[1][2].
- Integration Ecosystem: Compatible with QwenCode, Claude Code, Cline, and OpenClaw (formerly Moltbot/Clawdbot) development tools[3].
🔮 前景展望AI analysis grounded in cited sources
Multi-model switching reduces vendor lock-in for enterprise coding workflows
Developers can freely switch between Qwen, GLM, MiniMax, and Kimi models within a single subscription, enabling cost optimization and model selection based on task-specific performance rather than platform constraints.
Open-weight Qwen3-Coder-Next enables offline coding agent deployment
As an open-weight model designed for local development, organizations can deploy coding agents on-premises or edge infrastructure, reducing latency and data residency concerns compared to cloud-only competitors.
⏳ 時間線
2025-09-23
Qwen3-max snapshot released; baseline for subsequent thinking-mode improvements
2025-12-01
Qwen-plus-2025-12-01 model snapshot released as part of Qwen3 series
2026-01-23
Qwen3-max-2026-01-23 stable release with integrated thinking mode and tool-use capabilities
2026-01-31
Coding Plan subscription service activated on Alibaba Cloud Model Studio
2026-02-15
Qwen3.5-plus-2026-02-15 snapshot released with thinking mode enabled by default
2026-02-24
Alibaba Cloud Model Studio documentation updated to reflect full Coding Plan feature set and supported models
📎 來源 (8)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： 机器之心 ↗
每週 AI 簡報
每週一封，可隨時退訂。