--- type: "post" id: "sakana-ai-and-uc-berkeley-released-mass-a-def3df" slug: "sakana-ai-and-uc-berkeley-released-mass-a-def3df" title: "Sakana AI and UC Berkeley released MASS , a recursive self-improvement loop using one shared model for every role." description: "MASS treats a multi-agent workflow as something a model can both improve and learn from." retrieval_nugget: "MASS treats a multi-agent workflow as something a model can both improve and learn from." published_at: "2026-10-11" updated_at: "2026-10-11" record_date: "2026-10-11" date_kind: "published_at" topics: [] source_urls: ["https://pub.sakana.ai/mass"] schema_version: "newruntime-agent-readable-v0.2" stable_id: "post:sakana-ai-and-uc-berkeley-released-mass-a-def3df" status: "published" visuals: [] editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:59079482c62dff30d011f9d375505f83d926828ad9e8923873129c23fe573a41","reviewed_at":"2026-10-11T23:01:22.553Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0} analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"MASS makes the team’s accumulated work a self-supervision signal for a shared model.","observed_facts":[{"text":"The source says one shared language model serves as task solver, workflow optimizer, and evaluator.","source_urls":["https://pub.sakana.ai/mass"]},{"text":"The source describes an inner workflow-search loop with fixed weights and an outer loop that trains on selected workflow executions.","source_urls":["https://pub.sakana.ai/mass"]},{"text":"The source reports improved score per output token over the base model on research benchmarks and says external judgments were withheld from MASS.","source_urls":["https://pub.sakana.ai/mass"]}],"mechanism":"An inner loop has the model revise and judge multi-agent workflows; an outer loop distills selected executions into the shared model’s weights, which then resumes all three roles.","why_now":"The source motivates homogeneous recursive self-improvement for settings where expert assessment is scarce or slow, and reports benchmark gains from the proposed method.","implications":["Workflow search and execution traces can become training material rather than disposable orchestration.","The usefulness of the loop depends on whether its self-evaluator selects work that tracks real task quality."],"evidence_boundary":"The training tasks are non-verifiable and the self-evaluation signal’s correctness is unknown; external GPT-5.5 and Claude Opus 4.8 judgments were withheld from MASS and do not establish general reliability.","watch_conditions":["Whether gains persist on independently verified tasks or under external evaluation across repeated runs."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied for this specific MASS method; the source supports a distinct research record about workflow-derived self-supervision."} editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/posts/sakana-ai-and-uc-berkeley-released-mass-a-def3df/","markdown":"https://newruntime.com/posts/sakana-ai-and-uc-berkeley-released-mass-a-def3df.md","json":"https://newruntime.com/posts/sakana-ai-and-uc-berkeley-released-mass-a-def3df.json"} --- # Sakana AI and UC Berkeley released MASS , a recursive self-improvement loop using one shared model for every role. ## Retrieval answer MASS treats a multi-agent workflow as something a model can both improve and learn from. MASS treats a multi-agent workflow as something a model can both improve and learn from. Thesis: MASS makes the team’s accumulated work a self-supervision signal for a shared model. The public source says the same model acts as task solver, workflow optimizer, and evaluator; the optimizer revises how agents divide work, exchange results, and check one another, while the executor’s selected runs later supply training examples. Mechanism: An inner loop has the model revise and judge multi-agent workflows; an outer loop distills selected executions into the shared model’s weights, which then resumes all three roles. Why now: The source motivates homogeneous recursive self-improvement for settings where expert assessment is scarce or slow, and reports benchmark gains from the proposed method. That makes the design relevant to teams exploring improvement loops where a stronger teacher is unavailable. Evidence boundary: The training tasks are non-verifiable and the self-evaluation signal’s correctness is unknown; external GPT-5.5 and Claude Opus 4.8 judgments were withheld from MASS and do not establish general reliability. First implication: Workflow search and execution traces can become training material rather than disposable orchestration. First watch condition: Whether gains persist on independently verified tasks or under external evaluation across repeated runs. The source establishes a method and reports benchmark results, not general-purpose recursive self-improvement. Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis. - collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)