CHINA DEV RADAR
China Dev Radar #003 — Verified AI Engineering, Linux Updates and Developer Tools
Six evidence-reviewed stories from November 2–6, 2026: Atrex supervised GPU optimization, Meituan LongCat agent-training engineering, Baidu ERNIE base-model weights, openEuler LTS maintenance, Tencent Hunyuan Hy4 visual computer-use announcement, and deepin desktop engineering research. Hy4 weights were forthcoming at announcement. The deepin report is a November-published retrospective of October work, not a new November release.
Featured Developments
01
Atrex Kernel Agent 0.4.0 introduces supervised GPU kernel optimization
Originally published
What changed
Atrex 0.4.0 introduces the Light workflow and modular provider architecture for GPU-kernel optimization with coding agents. An independent Supervisor controls correctness tests, source snapshots, same-allocation ABBA latency comparisons and candidate promotion. The release adds native CUDA manifests, multi-file sources and improved recovery, while migration from v0.3.0 requires workflow and workspace changes. Performance and token-savings claims remain publisher-reported.
Sources
02
Meituan LongCat-2.5 details verifiable agent training and multimodal infrastructure
Originally published
What changed
Meituan's November 2 report describes a natively multimodal LongCat-2.5 model for long-horizon tasks, including complex codebase development. Its engineering account covers executable coding-task construction, fail-to-pass and pass-to-pass tests, asynchronous partial reinforcement-learning rollouts, and accelerator-aware long-context training and inference. Performance comparisons and throughput improvements are Meituan's own measurements, not independent benchmarks. This is distinct from the Meituan AI Coding article covered in Radar #002.
Related entity
Meituan →Sources
03
Baidu releases ERNIE 5.5 Flash Base model weights and hybrid attention design
Originally published
What changed
Baidu released FP8 and BF16 pretrained base weights for ERNIE 5.5 Flash Base, a sparse MoE model with 280B total and about 14B active parameters. Its original architecture report describes compressed attention, selective full-context layers, and SGLang/vLLM-serving pathways. The 128K training context is distinct from configurations supporting up to 1M tokens in inference. This is a base checkpoint, not an instruction-tuned chat model, and the weights use a dedicated ERNIE model license. Benchmark comparisons are publisher-provided.
Related entity
Baidu →Sources
Also Worth Knowing
04
openEuler issues November maintenance and security updates for LTS branches
Originally published
openEuler's Release, QA and CI/CD groups announced the update_20261102 train for 22.03-LTS-SP4 and multiple 24.03 LTS service-pack branches. The original announcement lists package changes, security advisories and branch-specific issues. Operators should check their exact branch and individual CVE bulletins before deployment. Publisher-reported vulnerability fix counts across branches must not be added as unique CVEs.
Sources
05
Tencent announces Hunyuan Hy4 vision and computer-use capabilities
Originally published
Tencent announced Hunyuan Hy4 on November 4 with visual understanding and computer-use tools for graphical and command-line workflows. The company described API availability and planned to release downloadable weights and a system card later. This is an announcement of model capabilities and hosted access, not evidence that open weights had already been released. Performance and safety assessments remain publisher-reported.
Related entity
Hunyuan →Sources
From the Research Desk
Earlier technical highlights, shown with their original publication dates.
06
Worth exploring: deepin's October Linux and desktop engineering report
Originally published
deepin's November 5 retrospective documents October work on device drivers, LoongArch compatibility, kernel maintenance, reproducible builds, Treeland compositor integration and DTK toolkit improvements. It is a useful technical map for Linux maintainers and desktop developers, not a new November kernel release. Radar #002 already covered the Linux 7.3 community preview.