Alibaba’s Qwen team introduced QwenGyre, an elastic reinforcement-learning framework for training coding agents on extreme long-horizon tasks through unmodified black-box harnesses. It moves idle GPUs between rollout and training and compresses branching agent histories into at most five trajectories per execution. On Qwen 3.6 122B and Qwen 3.8 2.4T, it was 1.21x to 1.85x faster than Colocate and up to 1.78x faster than Async at matched scores. A 48-step Qwen 3.8 2.4T run finished in 75.42 hours and lifted NL2RepoBench from 52.48% to 58.54%; the summary links no public code.
QwenGyre is a training framework, not a newly released model.
It trains agents via black-box harnesses on hour-long, near-1M-token rollouts.
Runs used Qwen 3.6 122B and flagship Qwen 3.8 2.4T.
Best reported speedup is 1.85x versus Colocate on TerminalBench.
NL2RepoBench rose from about 52.5% to 58.5%; no public code is linked.
Alibaba’s Qwen team has introduced QwenGyre, an elastic reinforcement learning RL framework for extreme-long-horizon agent training. QwenGyre is an online reinforcement learning system from Alibaba that trains coding agents through unmodified, black-box harnesses. It targets rollouts that run for hours and process close to 1M tokens. The QwenGyre elastic RL framework was used on Qwen’s flagship Qwen 3.8 2.4T model. It rests on 2 core ideas. It moves idle GPUs between rollout and training. It also turns branching agent histories into bounded training samples.
TL;DR
Size: A training framework, not a model. Tested on Qwen 3.6 122B and Qwen 3.8 2.4T (A95B per Qwen), with ~700K tokens per rollout.
Runs on: Multi-node GPU clusters. The 122B runs used 24 to 32 nodes; the 2.4T run used 192 nodes (4 cells of 48). GPU model not disclosed. Megatron trains, SGLang serves rollouts.
Performance: 1.38x to 1.78x faster than Async, 1.21x to 1.85x faster than Colocate, at matched scores.
Best: 1.85x end-to-end speedup over Colocate on TerminalBench.
Worst: Only 1.21x over Colocate on the 2.4T model, where many queries ran near the timeout.
Bottom line:
Best: NL2RepoBench rises from 52.5% to 58.5% in 48 steps.
Worst: the paper links no public code release.
What problem does QwenGyre solve?
The research team defines ‘XLong-horizon’ tasks as agent runs spanning hours and hundreds of model and tool interactions. Examples include generating full repositories from specs, as in NL2Repo-Bench.
Two problems appear at this scale:
First, rollouts have long tails. In a Colocate setup, training waits for the slowest execution while GPUs idle. In an Async setup, fixed GPU pools cannot lend spare capacity to each other.
Second, harnesses like Claude Code compact context, spawn sub-agents and retry. One execution becomes a branching tree, not a single trajectory. Expanding every path duplicates shared prefixes and inflates training cost.
How does the Elastic Scheduler work?
QwenGyre splits GPUs into cells that share a training layout. Each cell can serve rollouts or host a full training replica. A core cell owns the authoritative weights and optimizer state. Satellite cells join training later.
The scheduler tracks a rollout waterlevel: queued and running executions. As executions finish, cells switch to training once the remaining cells can still serve all live work. Model calls are rerouted through a proxy, so the harness keeps running.
Key mechanics:
KV-cache blocks migrate over RDMA during rerouting, reducing prefix recomputation.
Satellites pull core weights through Mooncake’s Transfer Engine and join an ongoing batch.
Streaming training lets cells claim micro-steps as data becomes ready.
Average role switches took 8.52 s (rollout to training) and 3.46 s (training to rollout).
How does the Trajectory Processor build training samples?
A proxy records exact token IDs and behavior log-probabilities using token-in, token-out (TITO). Calls are stored as prefix-shared trajectory trees.
Partial scoring: Timed-out executions are still scored on their preserved workspace. Evaluator failures are excluded, never treated as a zero reward.
Bounded sampling: At most 5 trajectories per execution are kept, ranked main > main-summary > sub-agent > sub-agent-summary.
Deduplication: Shared targets count once, and only policy-generated tokens are trained.
Fair weighting: Losses average within each execution, so extra branches do not inflate its weight.
A cap of 5 matched uncapped training scores while cutting forward-backward time by 25.2%. Training uses an adapted GSPO with 16 rollouts per group.
How fast is QwenGyre? Benchmark results
All runs used Claude Code 2.1.220 as the harness, with equal GPU budgets and matched staleness.
Workload (model)
vs Async
vs Colocate
NL2RepoBench (Qwen 3.6 122B, 48 steps)
1.42x to 1.53x
1.36x to 1.47x
DeepSWE (Qwen 3.6 122B, 24 steps)
1.569x to 1.572x
1.610x to 1.824x
TerminalBench (Qwen 3.6 122B, 24 steps)
1.383x to 1.430x
1.809x to 1.849x
NL2RepoBench (Qwen 3.8 2.4T, 48 steps)
1.78x
1.21x
On the 2.4T model, QwenGyre finished 48 steps in 75.42 hours. Async took 134.55 hours and Colocate took 91.47 hours. Evaluation pass rate rose from 52.48% to 58.54%.
How does QwenGyre compare with other agentic RL frameworks?
Asif Razzaq is the CEO of Marktechpost AI Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.
News Hub
Add as a preferred source on Google
