J-Comm Receiver Only - ALFWorld: leaderboard

J-Comm paper v1 Receiver-only baseline: single frozen Qwen3-4B, without sender/latent communication. bfloat16,vLLM0.10.2V0,temperature0.6,top-k20,top-p0.95. Three runs. Matched task-specific output budgets; truncated answers count incorrect. Multi-agent method and heterogeneous sender/receiver scores are not attributed to this model. 134unseentasks,2048tokens/turn,50turnlimit. Metric: Accuracy (%).

Metric: Accuracy (%). Source: arxiv.org. 1 models tracked.

Top models

#ModelScoreOverall rank
1Qwen3-4B49.25

Interactive version: theaggregate.ai/benchmark?slug=j-comm-receiver-only-alfworld · How It Works · Data refreshed daily, snapshot 2026-10-27.