routing-video-production-request-ordered-chain-registered-sub-agents-8b-open CLM-0480For routing a video-production request to an ordered chain of registered sub-agents, an 8B open model fine-tuned on descriptor-conditioned routing plans (SFT on verified plans plus GRPO rewarding validity, sub-agent coverage F1 and dependency-consistent order) picks chains more accurately than much larger prompted planners, with the gap widest on long, structurally unfamiliar chains.
Capability: Getting more from several models than from one
Observed on
Sequential plans over a fixed inventory of 20 sub-agents, each appearing at most once; evaluated as chain/step match against reference plans, not as end-to-end video quality..
Sources
- Validated by: Exact-match chain accuracy, per-step accuracy and slot-aware edit distance on two held-out sets (DCE 522 easy, DCH 200 hard): 91.0/81.5 chain accuracy versus Gemini 3 Pro at 80.3/47.5, GPT-4o 58.0/31.0, and Qwen3-8B base 29.5/11.5; ablations show GRPO adds +33.5 chain points on DCH over the SFT checkpoint, and the frozen Director selected two post-training sub-agents correctly in 93.0% and 99.2% of cases.. Measured on 722 held-out routing cases with matched prompts and no few-shot examples, plus 4B/8B and supervision-format ablations. [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In multi-agent LLM pipelines over reasoning and code tasks with a pool of 3B-32B open-weight backbones, conditioning each per-step routing decision on a learned filtered memory rather than the full accumulated trajectory leaves accuracy about the same while cutting inference cost by roughly 40-50%, so most of the accumulated history is cost without accuracy value.Getting more from several models than from one · reviewed by AI
- In multi-hop KGQA evaluated as graph navigation with annotated gold evidence paths, reported agreement between an agent's traversed path and the reference path cannot be credited to question conditioning unless it is compared against question-agnostic references — unbiased random walks and answer-oracle shortest paths — because dense or small graphs let structure alone deliver both answer reachability and partial path overlap.Whether the measurement made the finding · reviewed by AI
- In multi-agent LLM workflows with a mixed-size model zoo, choosing which model instantiates each agent step-by-step from an online estimate of task-progress gain plus remaining time and cost budget keeps task quality near the best single-model policy at lower energy than one-shot query-level routing or stall-triggered escalation.Getting more from several models than from one · reviewed by AI
- For long-horizon embodied task planning where a symbolic model defines action preconditions and effects, correcting a violated plan by edits the action model itself implies (inserting the missing navigate or open) raises plan success more per LLM call than returning the violation to the LLM for replanning, and the gain persists for frontier models as well as compact ones.Planning a sequence of actions before acting · reviewed by AI
- When RL-training a central-agent multi-agent harness with 1.5B-3B Qwen2.5 backbones on multi-hop QA, optimizing only the coordinating central policy and leaving sub-agents frozen scores higher than jointly training central and sub-agent policies (average F1 47.79 vs 45.36, accuracy 42.88 vs 40.49 at 3B).Getting more from several models than from one · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A prompted frontier planner matching or beating the fine-tuned 8B Director on long-chain routing on the same held-out sets, or the tuned Director's advantage disappearing on routing tasks with a different sub-agent inventory or non-sequential plans. Automatic check flagged: figures not in the source: 722.