๐งงQwen (GitHub Releases: qwen-code)โขStalecollected in 2h
Qwen-code v0.18.0-preview.0: CLI and Workflow Enhancements
๐กMajor CLI update for Qwen-code featuring qwen3.7-plus multimodal support and new automated developer workflows.
โก 30-Second TL;DR
What Changed
Added standalone auto-update support for the CLI.
Why It Matters
These updates streamline the development loop for engineers using Qwen-code by automating maintenance tasks and providing better visibility into model interactions. The addition of multimodal support for qwen3.7-plus expands the utility of the tool for complex coding tasks.
What To Do Next
Update your Qwen-code CLI to v0.18.0-preview.0 and test the new /fork background-agent command to optimize your multi-tasking workflow.
Who should care:Developers & AI Engineers
Key Points
- โขAdded standalone auto-update support for the CLI.
- โขImplemented multimodal support for qwen3.7-plus.
- โขIntroduced automated PR/issue triage workflows via @qwen/triage.
- โขEnhanced CLI productivity with new commands like /fork and /copy N.
- โขImproved UI/UX with model name display and skill picker dialogs.
๐ง Deep Insight
Web-grounded analysis with 20 cited sources.
๐ Enhanced Key Takeaways
- โขQwen-code is an open-source AI coding agent for the terminal, optimized for Qwen series models, and is a fork of Google's Gemini CLI, offering multi-protocol support for various LLM APIs including OpenAI, Anthropic, and Gemini-compatible endpoints.
- โขThe
qwen3.7-plusmodel, now supported by Qwen-code, functions as a multimodal interactive hybrid agent capable of perceiving real-world scenes, interpreting graphical user interfaces (GUIs), and generating code from visual inputs, extending beyond traditional text-based coding. - โขQwen-code facilitates a comprehensive agentic workflow through built-in 'Skills' and 'SubAgents,' enabling it to autonomously build features from descriptions, debug complex issues, navigate large codebases, and automate tedious tasks like fixing lint issues or resolving merge conflicts.
- โขIt offers flexible deployment options including an interactive terminal UI, a headless mode suitable for scripts and CI/CD pipelines, and direct integration with popular Integrated Development Environments (IDEs) such as VS Code, Zed, and JetBrains.
๐ Competitor Analysisโธ Show
| Feature/Aspect | Qwen-code (with Qwen3-Coder/Qwen3.7-Plus) | Claude Code | Gemini CLI | DeepSeek (LLM for coding) |
|---|---|---|---|---|
| Nature | Open-source CLI agent, optimized for Qwen models | Proprietary terminal AI assistant | CLI tool, access to Gemini models | Open-source LLM, excels in code generation |
| Multimodal Support | Yes (with Qwen3.7-Plus: text, image, video input, GUI interaction) | Limited/Text-based (focus on codebase understanding) | Yes (with Gemini models) | Primarily text-based for code |
| Context Window | Native 256K, extensible to 1M tokens (Qwen3-Coder); 1M tokens (Qwen3.7-Plus) | Leverages Claude 3.7 Sonnet's context capabilities (specific limits not detailed) | Access to Gemini 2.5 Pro's 1M token context window | Not specified for CLI, but LLM context varies. |
| Agentic Workflow | Rich built-in tools (Skills, SubAgents), automated PR/issue triage, multi-step automation | Deep codebase awareness, multi-file edits, integrates with build tools/test suites, explicit approval for changes | Can ground prompts with Google Search, supports MCP for extensibility, custom prompts, scripting | Focus on code generation, debugging, refactoring (less on agentic workflow orchestration) |
| IDE Integration | Optional integration for VS Code, Zed, JetBrains IDEs | Integrates with VS Code and JetBrains | Not explicitly mentioned for direct IDE integration beyond CLI usage. | N/A (LLM, not a CLI tool) |
| API Compatibility | OpenAI / Anthropic / Gemini-compatible APIs, Alibaba Cloud Coding Plan, OpenRouter, Fireworks AI | Anthropic API protocol | Gemini API | N/A (LLM, but can be accessed via APIs) |
| Pricing Model | Depends on deployment choice (API Key, Alibaba Cloud Coding Plan, other providers) | Subscription with pay-per-use characteristics | Generous free tier, then pay-per-use | N/A (open-source weights, API pricing varies by provider) |
๐ ๏ธ Technical Deep Dive
- Qwen models are built on a Transformer-based architecture, with some variants utilizing a Mixture-of-Experts (MoE) design, such as Qwen3-235B-A22B (235 billion total parameters with 22 billion activated) and Qwen3-Coder-480B-A35B (480 billion total with 35 billion active parameters).
- Qwen3 models incorporate a unique dual-mode capability, allowing seamless switching between a 'thinking mode' for complex, multi-step reasoning and a 'non-thinking mode' for rapid, context-driven responses, along with a 'thinking budget' mechanism for adaptive resource allocation.
- The Qwen3-Coder model natively supports a 256K token context window, which can be extended up to 1M tokens using techniques like YaRN, enabling deep, repository-wide code understanding.
- Qwen3.7-Plus employs a hybrid sparse mixture-of-experts architecture that combines Gated DeltaNet linear attention with standard gated attention layers for efficient inference, and features a 1-million token context window with up to 256K tokens specifically allocated for internal chain-of-thought processing.
- Qwen models are pre-trained on extensive and diverse multilingual datasets; for instance, Qwen3 was trained on 36 trillion tokens across 119 languages and dialects.
- Qwen-code itself is implemented in TypeScript and utilizes a specialized tool parser for robust function calling, ensuring consistency with Qwen3's updated special tokens and token IDs.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
Increased adoption of Qwen-code for complex, multi-modal software development tasks.
The integration of
qwen3.7-plus with its advanced multimodal capabilities (GUI interaction, visual code generation) and automated triage workflows will enable developers to handle more sophisticated and visually-driven development tasks directly from the CLI.Enhanced efficiency in developer workflows through deeper AI integration.
Auto-update, improved status displays, and expanded skill management will streamline the use of Qwen-code, reducing manual overhead and allowing developers to focus more on core logic rather than tool maintenance or configuration.
Broader competition in the AI coding agent market, particularly for open-source and multi-protocol solutions.
Qwen-code's open-source nature and multi-protocol support (OpenAI, Anthropic, Gemini-compatible APIs) position it as a flexible alternative, potentially driving other providers to offer similar interoperability and agentic features.
โณ Timeline
2023-04
Alibaba launched a beta of Qwen (Tongyi Qianwen).
2023-08
Alibaba released Qwen-7B as an open-weight model under Apache 2.0 license.
2024-06
Qwen2 series released, introducing major architectural revisions.
2025-05
Qwen3 model family released, featuring MoE architectures and thinking modes.
2025-07
Qwen-Code CLI tool released, optimized for Qwen3-Coder model.
2026-05-31
Qwen3.7-Plus, a multimodal interactive hybrid agent model, was introduced.
๐ Sources (20)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
๐ฐ
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Qwen (GitHub Releases: qwen-code) โ