QwenGyre

Headlines · 2

1. 3d
   Personal agent wave: Qwen Go opens invite testing as Meta ships Muse Wonder

   At its Yunqi conference in late September, Alibaba said Qwen would accelerate its push into personal agents; on October 13 it unveiled the personal assistant "Qwen Go" with limited invite testing in the Qwen app. ByteDance's Doubao is reported to be testing a personal agent project codenamed Spell and merging it with existing products, while Tencent has TX Handy Bot and Yuanbao Assistant; Meta's Muse is credited with igniting the wave, and OpenAI followed with dots. On October 12 Meta released the Muse Wonder series (Muse Wonder and Muse Wonder Full), centered on Active Memory and Active Focus, described as ranking just behind Opus 5.5 on an AA leaderboard, with internal MemoryBench and FocusBench tests for long-dialogue recall and sustained progress. Qwen's team also published QwenGyre, an online reinforcement learning framework for very long-horizon agent tasks, submitted to arXiv on September 27; it reports Qwen 3.8 2.4T lifting NL2RepoBench pass rate from about 52.5% to 58.5% after 48 RL steps and cutting training time roughly 44% versus an Async baseline at similar scores.

   36氪 人工智能 · 🔥 0
2. 5d
   Alibaba Qwen releases QwenGyre, an online RL framework for long-horizon agent tasks

   Alibaba's Qwen team released QwenGyre, an online reinforcement learning framework aimed at long-horizon agent tasks, on October 12. According to the company, such tasks can run for hours, involve hundreds of interdependent interactions with the environment and process millions of tokens; QwenGyre splits GPUs into compute units so that units freed when a batch of tasks finishes roll into training while running tasks keep their progress and caches, and it uses Token-in Token-out recording to organize shared-prefix trajectories into a tree, expanding only selected paths as training samples while Partial Scoring salvages feedback from timed-out but partially completed tasks. In the reported NL2RepoBench online RL experiment, with roughly 700,000 tokens per rollout, pass rate rose from 52.5% to 58.5% after 48 training steps, and QwenGyre delivered a 1.78x end-to-end speedup over an Async implementation at the same GPU budget, with cross-task end-to-end speedups on Qwen3.6-122B across NL2RepoBench, DeepSWE and TerminalBench.

   IT之家 AI · 🔥 0

Experience and discussion from the community

Share my QwenGyre experienceAsk about QwenGyre