{"paper":{"title":"AgentPSO: Evolving Agent Reasoning Skill via Multi-agent Particle Swarm Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"AgentPSO evolves multi-agent reasoning skills by treating each agent's natural-language description as a particle state that updates toward better performance.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Choonghan Kim, Hangeol Chang, Hyunmin Hwang, Jaemin Kim, Jong Chul Ye","submitted_at":"2026-05-09T05:38:21Z","abstract_excerpt":"Multi-agent reasoning has shown promise for improving the problem-solving ability of large language models by allowing multiple agents to explore diverse reasoning paths. However, most existing multi-agent methods rely on inference-time debate or aggregation, which can be vulnerable to incorrect peer influence and biased consensus. Moreover, the agents themselves remain static, as their underlying reasoning skills do not evolve across tasks. In this paper, we introduce \\textbf{AgentPSO}, a particle-swarm-inspired framework for evolving multi-agent reasoning skills. AgentPSO treats each agent a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on mathematical and general reasoning benchmarks show that AgentPSO improves over static single-agent skills and test-time-only multi-agent reasoning baselines. The evolved skills further transfer across benchmarks and to another backbone model, suggesting that AgentPSO captures reusable reasoning procedures rather than merely optimizing benchmark-specific prompts.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That iterative semantic updates to natural-language skill descriptions (combining previous velocity, personal-best, global-best, and self-reflective directions) produce genuine improvements in reasoning capability rather than superficial prompt changes, and that these updates can be performed without access to model internals or gradients.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AgentPSO evolves reusable multi-agent reasoning skills via PSO-inspired natural-language updates, outperforming static agents and test-time multi-agent baselines on math and general reasoning tasks with cross-benchmark transfer.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AgentPSO evolves multi-agent reasoning skills by treating each agent's natural-language description as a particle state that updates toward better performance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9ebb55e2df5cfc6a9888629b9d4df6526e3b06d68df2db55481025044793e443"},"source":{"id":"2605.08704","kind":"arxiv","version":2},"verdict":{"id":"4b12bc81-0956-402a-b85f-d97f2c8b31ed","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T00:51:04.278115Z","strongest_claim":"Experiments on mathematical and general reasoning benchmarks show that AgentPSO improves over static single-agent skills and test-time-only multi-agent reasoning baselines. The evolved skills further transfer across benchmarks and to another backbone model, suggesting that AgentPSO captures reusable reasoning procedures rather than merely optimizing benchmark-specific prompts.","one_line_summary":"AgentPSO evolves reusable multi-agent reasoning skills via PSO-inspired natural-language updates, outperforming static agents and test-time multi-agent baselines on math and general reasoning tasks with cross-benchmark transfer.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That iterative semantic updates to natural-language skill descriptions (combining previous velocity, personal-best, global-best, and self-reflective directions) produce genuine improvements in reasoning capability rather than superficial prompt changes, and that these updates can be performed without access to model internals or gradients.","pith_extraction_headline":"AgentPSO evolves multi-agent reasoning skills by treating each agent's natural-language description as a particle state that updates toward better performance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.08704/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T09:02:01.969011Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T22:35:59.838484Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T14:31:17.734347Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T10:51:40.305207Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"4bdd577a49c27af87811804832595b16738862e725cf41f7190098b4a18b5091"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}