{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2611.08483v1",
  "slug": "2611-08483v1-1jfv76o",
  "url": "https://feed7.dev/p/2611-08483v1-1jfv76o",
  "title": "SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization",
  "why_included": "SwarmEvolve uses latent communication between agent swarms to search for better programs. Its reported speed suggests a different way to structure expensive, iterative coding-agent experiments.",
  "summary": "SwarmEvolve runs **hundreds of low-cost agents per round** in swarms, which share KV caches and search in latent space. On **21 problems**, it reached scores matched by other frameworks in 60 minutes in a median of **3–8 minutes**.\nFor coding-agent builders, the useful idea is to diversify candidate solutions and combine them without asking every agent to write a long textual reasoning trace. The paper also reports that Qwen3 8B improved on some solutions first found by AlphaEvolve with frontier models.\nThe results concern mathematical discovery and scientific computing, not general software development. The material does not establish whether the approach transfers to everyday coding tasks or what infrastructure is needed to reproduce it.",
  "practical_implication": "",
  "agent_context": "SwarmEvolve runs **hundreds of low-cost agents per round** in swarms, which share KV caches and search in latent space. On **21 problems**, it reached scores matched by other frameworks in 60 minutes in a median of **3–8 minutes**.\nFor coding-agent builders, the useful idea is to diversify candidate solutions and combine them without asking every agent to write a long textual reasoning trace. The paper also reports that Qwen3 8B improved on some solutions first found by AlphaEvolve with frontier models.\nThe results concern mathematical discovery and scientific computing, not general software development. The material does not establish whether the approach transfers to everyday coding tasks or what infrastructure is needed to reproduce it.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2611.08483v1",
    "published_at": "2026-11-05T18:59:29.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "agent",
  "domains": [
    "coding",
    "research"
  ],
  "topics": [
    "multi-agent",
    "harness-engineering"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [],
  "connected_context": {
    "meaning": "SwarmEvolve adds evidence for low-cost parallel search that shares KV caches and avoids long textual reasoning traces, with reported speedups on mathematical and scientific-computing problems. It strengthens the case for testing diverse candidate solutions, but the supplied results do not show transfer to everyday software development or provide enough infrastructure detail to reproduce them.",
    "corpus_size": 1020,
    "generated_at": "2026-11-06T18:41:16.770Z",
    "connections": [
      {
        "title": "HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.15938v1",
        "feed7_url": "https://feed7.dev/p/2609-15938v1-0oezpu6",
        "reason": "Both study multi-agent scientific search, but HypoEvolve formalizes population updates through a genetic algorithm while SwarmEvolve searches in latent space and shares KV caches."
      },
      {
        "title": "Stellar Colosseum: A Many-Agent Harness for Long-Horizon Research in Mathematics and Theoretical Computer Science",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.15983v1",
        "feed7_url": "https://feed7.dev/p/2609-15983v1-13dydbk",
        "reason": "Both use many-agent search in mathematical research; Stellar Colosseum emphasizes strategy exploration, falsification, and verifier feedback, while SwarmEvolve reports latent-space search and faster matched scores on its stated problems."
      },
      {
        "title": "Agent swarms and the new model economics",
        "source_name": "Cursor",
        "source_url": "https://cursor.com/blog/agent-swarm-model-economics",
        "feed7_url": "https://feed7.dev/p/agent-swarm-model-economics-02n8muc",
        "reason": "Both concern swarm efficiency, but Cursor's SQLite experiment centers on planner-worker context separation and coordination churn, whereas SwarmEvolve's reported mechanism includes shared KV caches and latent-space search."
      },
      {
        "title": "alphaXiv/OpenResearch",
        "source_name": "GitHub",
        "source_url": "https://github.com/alphaXiv/OpenResearch",
        "feed7_url": "https://feed7.dev/p/openresearch-1472dzs",
        "reason": "SwarmEvolve reports results from parallel scientific search, while OpenResearch offers isolated worktrees and experiment lineage for reproducible parallel research; it supplies workflow context for evaluating such a search approach, not evidence that its results transfer."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-11-05T18:59:29.000Z",
  "modified_at": "2026-11-05T18:59:29.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2611-08483v1-1jfv76o",
    "json": "https://feed7.dev/p/2611-08483v1-1jfv76o.json",
    "markdown": "https://feed7.dev/p/2611-08483v1-1jfv76o.md"
  }
}