https://schristoph.online/tags/harnessengineering/Personal homepage and blog of Stefan ChristophHugo -- gohugo.ioen-usStefan Christoph. All rights reserved.Mon, 08 Jun 2026 00:00:00 +0000https://schristoph.online/blog/cheapest-model-writes-the-harness/?utm=rss-feedMon, 08 Jun 2026 00:00:00 +0000https://schristoph.online/blog/cheapest-model-writes-the-harness/<blockquote> <p><strong>TL;DR:</strong> A May 2026 paper separates two capabilities that self-improving agents usually conflate: <em>writing</em> harness updates and <em>benefiting</em> from them. Writing is flat across model tiers: a 9B open model produces updates roughly as useful as a frontier model. Benefiting is an inverted-U, where mid-tier models gain most, strong models sit near the ceiling, and weak models can&rsquo;t follow the harness they&rsquo;re given. The practical move: put your cheap model in the evolver seat and your expensive model in the solver seat. I reproduced the mechanism on Amazon Bedrock, where a Haiku-written skill lifted a Sonnet solver from fail to pass.</p>