--- format: "aidr-story-markdown/v1" id: "77744cb30da4f3b58e25980759b18461bb411ad34faea27acc5a48af896dd5de" canonical_url: "https://aidr.today/77744cb3?lang=en" title: "Sakana AI's New MASS Method Boosts 27B Model Performance 1.6x" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-10-11T22:10:14.000Z" category: "Agents" topics: ["agent"] source_urls: ["https://huggingnews.com/ai/sakana-ais-new-mass-method-boosts-27b-model-performance-16x-2eb713c4","https://x.com/SakanaAILabs/status/2109385006603006435","https://x.com/hardmaru/status/2109385496237678844","https://x.com/nptacek/status/2109401063526936747","https://marketbrief.now/ai/sakana-ais-new-mass-method-boosts-27b-model-performance-16x-2eb713c4"] summary: "Sakana AI and UC Berkeley developed the Multi Agent Self Supervision (MASS) method, which uses virtual subagents to train a 27B model, increasing scores per output token by 1.2-1.6x across four research benchmarks. The approach addresses a supervision bottleneck in complex tasks but raises safety concerns about models reinforcing their own blind spots." --- # Sakana AI's New MASS Method Boosts 27B Model Performance 1\.6x > [Open the canonical story]() **Published:** 2026-10-11T22:10:14.000Z **Category:** Agents **Topics:** agent ## Summary Sakana AI and UC Berkeley developed the Multi Agent Self Supervision \(MASS\) method, which uses virtual subagents to train a 27B model, increasing scores per output token by 1\.2\-1\.6x across four research benchmarks\. The approach addresses a supervision bottleneck in complex tasks but raises safety concerns about models reinforcing their own blind spots\. ## Sources - [Story source]() - [Story source]() - [Supporting source]() - [Supporting source]() - [Story source]()