Alibaba merges Qoder's agent harness into open-source Qwen Code 1.0 Preview under Apache-2.0
Alibaba announced Qwen Code 1.0 on November 11, merging the agent harness from its Qoder coding product into the open-source Qwen Code project. The preview sits on a separate branch of QwenLM/qwen-code, and Alibaba says developers can now embed the harness in their own products.
Alibaba announced Qwen Code 1.0 on November 11, merging the agent "harness" built in its Qoder coding product into the open-source Qwen Code project. The harness is the layer that connects a model, its tools and the execution process. Alibaba says developers can now embed it in their own products.
Release details
Qwen Code 1.0 Preview is available on a separate branch of the QwenLM/qwen-code repository. The existing stable branch stays in place. GitHub lists a v1.0.0-preview pre-release dated November 11 alongside the current 0.27 line, and shows the project under the Apache-2.0 license.
Qwen Code is an open-source coding agent that runs in the terminal. The repository also shows desktop, browser, editor and chat integrations, including VS Code, Zed and JetBrains.
Architecture changes
According to Alibaba's description, relayed by IT Home (a Chinese-language outlet) and reported by Pandaily, version 1.0 starts from an in-process, in-memory event bus and extends a unified event model toward distributed operation.
- Intake, scheduling and execution are layered.
- A "brain and hand separation" keeps reasoning orchestration and tool execution under independent management.
- The stated aim is to move from one process hosting several sessions to multiple nodes hosting many more, while keeping the existing entry points and return paths.
Four practical effects Alibaba lists
- Pluggable processing units. Units for reasoning, memory and tools can be plugged in around events with less intrusion into the main execution flow.
- Single model entry point. Model access goes through one entry point that centralizes configuration, authorization and call records.
- Sandbox reuse. Sandbox resources are allocated and reused within the same configuration and permission boundaries. Alibaba says this should reduce repeated initialization and idle usage, and so lower running costs.
- Unified MCP management. MCP, the Model Context Protocol for connecting tools, is managed in one place for configuration, authorization and connections.
What is not yet available
Alibaba has not published benchmark results or a release date for the stable 1.0 version. The announcement does not say how closely the open-source kernel matches what runs inside the Qoder product. The "preview" label suggests interfaces may still change. The performance and cost-saving effects above are Alibaba's own statements, and no independent tests have been published.
No model is released with this update, and pricing is not applicable to the open-source project.
What this means
The release makes the execution layer of a commercial coding product available under a permissive license. Teams building their own agent products can inspect and embed that layer instead of writing the orchestration, model access and tool management themselves.
The design points to where Alibaba is putting its effort: separating reasoning from tool execution, and moving from single-process to multi-node operation. Centralized model access and MCP management also put configuration, authorization and call records in one place, which matters for teams that need to control how agents reach models and tools.
The caveats are material. The cost claim for sandbox reuse is unverified, and the preview branch carries no stability guarantee. Until Alibaba clarifies how the open-source kernel relates to the Qoder product, developers cannot assume the two behave identically. The stable 0.27 line remains the supported option for anyone who needs predictable interfaces today.
Related Articles
TIER IV Open-Sources Five End-to-End Autonomous-Driving AI Models for Autoware
TIER IV announced on November 11, 2026 the development and open-source availability of five end-to-end autonomous-driving AI models for Autoware. The lineup covers camera-only systems and configurations that add LiDAR, radar and high-definition maps.
Google ships Gemini Notebook API in public preview: REST and MCP server, source ingestion only
Google has released a public-preview REST API and MCP server for Gemini Notebook (formerly NotebookLM) that work with personal Google accounts. They can create notebooks and add web, YouTube, text and Markdown sources, but cannot yet query notebooks or generate outputs.
Adobe previews MCP access letting third-party AI agents edit Photoshop, Premiere files
At Adobe Max in Miami, Adobe previewed local Model Context Protocol integrations that let agents from OpenAI, Google, and Anthropic edit native project files in Photoshop, Illustrator, Premiere, and After Effects. A private beta is planned for early 2027. Adobe also added Plan Mode and a Model Selector to the Firefly AI Assistant beta.
Adobe adds planning mode and preview of always-on recurring-task agents to Firefly AI Assistant
Adobe announced an updated Firefly AI Assistant at its Max conference in Miami on November 10, 2026. It adds a planning mode for reviewing multi-step workflows and a preview of always-on agents for recurring tasks. Adobe is also bringing an agent to the Premiere mobile app and testing MCP connections for Photoshop, Illustrator, Premiere and After Effects.
Comments
Loading...