Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Qwen Code 0.26.9 hardens the Skills system
Qwen Code released 0.26.9 with a hardened Skills system: skill ZIP uploads, daemon APIs, TypeScript SDK support, bulk activation, and recovery protections make installing and managing Skills more reliable.
- Qwen Code 0.26.7 adds operator-scoped catalog refresh controls
Qwen Code released 0.26.7, adding an operator-scoped model.catalogRefresh setting and letting privacy.usageStatisticsEnabled: false stop background catalog refreshes while local metadata remains available.
- Qwen Code releases v0.26.5 with bounded SQL history queries
QwenLM published qwen-code v0.26.5 on GitHub. The release adds bounded SQL history queries for Managed local H2 and Hosted MySQL (paged recent messages, journal transactions, exact operation/command lookups) and lists no known breaking changes.
- Qwen Code v0.26.3 improves tool-call recovery, memory extraction, and qwen context-window detection
QwenLM’s qwen-code v0.26.3 has no known breaking changes. It improves tool-call recovery (including outer XML calls with quoted content), adds opt-in memory extraction cadence after a no-op run, shrinks tool results to the headroom left before auto-compaction, and treats invite-only qwen-latest-series aliases and qwen4 variants as high-context models.
- Qwen Code v0.26.1 adds daemon token persistence and Managed Session recovery upgrades
QwenLM tags qwen-code v0.26.1 with opt-in daemon token persistence, idle-only manual context compression for Workspace-bound Managed Sessions, OpenTUI input parity with Ink, and a qqbot breaking fix that restores per-group session isolation under thread scope.
- Qwen Code TypeScript SDK v0.1.22 bundles Qwen Code CLI 0.26.0
The qwen-code TypeScript SDK release v0.1.22 ships with version 0.26.0 of the Qwen Code command-line tool, built from the same source branch as the SDK.
- Qwen introduces QwenGyre, an elastic RL framework for training agents on extremely long tasks
Qwen's QwenGyre paper describes an online reinforcement learning framework for agents whose single runs span hours and close to 1M tokens. An elastic scheduler moves GPUs between rollout and training without interrupting live executions, and a trajectory processor turns branching runs into deduplicated, prefix-shared trees and scores partial progress. Scaled to Qwen 3.8 2.4T with about 700K tokens per rollout, it lifted NL2RepoBench from 52.5% to 58.5% in 48 steps, with up to 1.85x and 1.78x speedups over Colocate and Async training.
- Qwen releases Qwen-Image-2.1-Turbo, an 8-step accelerated checkpoint for image generation and editing
Qwen published Qwen-Image-2.1-Turbo on Hugging Face, an accelerated checkpoint of Qwen-Image-2.1 for text-to-image generation and image editing in 8 denoising steps. It uses the same 7B visual generation architecture, loads with QwenImage21Pipeline in Diffusers, ships its recommended sampling schedule, and reuses text and reference-image context across steps via prefix KV caching.
- Alibaba Cloud teases a Qwen-generated scene and points to Cloud and Model Studio APIs
Alibaba Cloud posted a teaser that a Qwen-generated scene looks unlike anything previously seen in reality, with links to Qwen Cloud and Model Studio API access.
- Ignite Vision builds Qwen-powered influencer matching that cuts screening to minutes
Alibaba Cloud said Ignite Vision built AI creator matching on Qwen that cuts influencer screening from hours of manual work to minutes across millions of KOL profiles.
- Qwen Code TypeScript SDK 0.1.18 bundles Qwen Code CLI 0.25.0
Alibaba's Qwen team released version 0.1.18 of the Qwen Code TypeScript SDK, which now ships with Qwen Code CLI 0.25.0 built from the same source branch. The release notes list no other changes.
- qwen-image-2.1-pro appears in Alibaba's model catalog without official announcement
Alibaba's image model qwen-image-2.1-pro has appeared in the Alibaba Model Studio catalog, with no separate official announcement yet. It is distinct from the 7B Qwen-Image-2.1 open-sourced on September 20, and a preview build (pre-qwen-image-2.1-pro-yunqi) had been seen in public use cases in late September.