AI
CompoWorld: Compositional Environment Scaling for General Agents
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 26/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 03:21:41 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 03:18:25 UTC.
Briefings
CompoWorld introduces Compositional Environment Scaling for general agents by composing a library of reusable services to expand task spaces beyond single environments. The system constructs 448 services exposing 10,130 tools, enabling agents to connect information across services via a verified dependency graph. Training uses 3K supervised fine-tuning trajectories and 1K reinforcement learning tasks on Qwen3.6-35B-A3B, resulting in a 9.17 point average improvement across eight benchmarks, including surpassing Claude Opus 4.6 on AutomationBench.
Key points
- CompoWorld constructs 448 services exposing 10,130 tools to enable multi-service information flow.
- The system uses 3K supervised fine-tuning trajectories and 1K reinforcement learning tasks to train Qwen3.6-35B-A3B.
Context and limitations
Results are based on a single preprint from September 2026; experimental data includes only 3K SFT and 1K RL trajectories, and benchmark comparisons are limited to eight specific metrics.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 03:21:41 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeid3zwkw6t54yksbxiengwxvyddq33sel6hztf2gtkp2i2havz5pcu