GLM5.3

Headlines · 3

1. 2d
   Doubao Work faces workflow and adoption hurdles in enterprise push

   A New Berry Daybreak investigation, authorized for republication by 36Kr AI and Cyzone Tech, examines ByteDance’s enterprise push for Doubao Work, including its Feishu integration, seat-plus-usage pricing, sales reorganization and announced opening to third-party models including Kimi 3.1 Pro, GLM5.3 and DeepSeek V4.1 Flash. User examples indicate that Doubao Work can save time on approvals, analysis and reporting, but may omit conditional steps or use outdated information. WorkBuddy, Qwen Office and Baidu Dazi are pursuing enterprise markets through different products and industries. ByteDance has not disclosed Doubao Work’s sales or contract figures.

   36氪 人工智能 · 🔥 10
2. Oct 20
   METASTONE launches TokenFleet inference cluster scheduling framework

   At the 2026 CCF National High Performance Computing conference, METASTONE (是石科技) released TokenFleet, a distributed scheduling framework focused on request routing, cross-datacenter capacity coordination and mixed-workload governance. The framework combines TokenPilot for instance selection, FleetMind for cross-datacenter capacity coordination and SLOMax for admission, queue governance and capacity allocation by service tier. METASTONE says that on its GLM5.3 tuning data, under the tested GPU cluster and workload, predictive-latency routing cut P95 time-to-first-token by 32.4% and P50 by 71.3%, while an SLOMax example estimated about 29.4% higher goodput; the report notes these are vendor measurements and that lower latency does not directly mean lower cost.

   量子位（原生 RSS） · 🔥 0
3. Oct 9
   PaperBenchX: best AI agent fully reproduces only 14% of papers

   UniPat AI's PaperBenchX benchmark gives 10 frontier agent configurations a published paper and asks them to re-run the key experiments in real scientific software, across 93 tasks covering 12 research directions. According to the team's published results, 70 of the 93 tasks were not fully reproduced by any configuration, and the top performer, GPT-6 Astra, fully cleared only 14% (13 tasks), while Chinese models clustered around 7% — roughly half that. Scoring separates modeling, execution and scientific verification across 3,168 scorable criteria, with a 4–24 hour per-task budget and outputs wiped and re-run in an isolated environment before grading.

   品玩 实时要闻 · 🔥 0

Experience and discussion from the community

Share my GLM5.3 experienceAsk about GLM5.3