This is an interesting direction for evaluating how well LLMs can actually improve agent systems, not just solve individual tasks. The focus on held-out evaluation and stochastic, budgeted optimization makes the benchmark especially useful for measuring genuine harness improvement rather than overfitting.</p>\n","updatedAt":"2026-08-07T06:42:15.606Z","author":{"_id":"6a757d4c0015bed8b66cbaac","avatarUrl":"/avatars/3c68d4d41d7620b3d926bcd987073629.svg","fullname":"say ana","name":"shay049ana","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9272206425666809},"editors":["shay049ana"],"editorAvatarUrls":["/avatars/3c68d4d41d7620b3d926bcd987073629.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.06301","authors":[{"_id":"6a753b60e1228e04b3238146","name":"Varun Ursekar","hidden":false},{"_id":"6a753b60e1228e04b3238147","name":"Apaar Shanker","hidden":false},{"_id":"6a753b60e1228e04b3238148","name":"Yash Maurya","hidden":false},{"_id":"6a753b60e1228e04b3238149","name":"Shehab Yasser","hidden":false},{"_id":"6a753b60e1228e04b323814a","name":"Vijay S. Kalmath","hidden":false},{"_id":"6a753b60e1228e04b323814b","name":"Veronica Chatrath","hidden":false},{"_id":"6a753b60e1228e04b323814c","name":"Yuan Xue","hidden":false}],"publishedAt":"2026-08-06T00:00:00.000Z","submittedOnDailyAt":"2026-08-07T00:00:00.000Z","title":"HarnessOpt-Bench: Evaluating LLMs at Harness Optimization","submittedOnDailyBy":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user","name":"taesiri"},"summary":"As LLMs are increasingly deployed within agentic systems, their capabilities depend not only on the model weights but also on the harness: the prompts, tools, control flow, memory, and orchestration code surrounding them. This makes automated harness optimization -- the iterative and evaluation-guided improvement of a harness by an AI system -- both an important route to improving AI systems and a demanding capability for AI systems themselves. Yet the community lacks a common protocol for measuring how well frontier LLMs perform at this task. We introduce HarnessOpt-Bench, a benchmark for end-to-end harness optimization under expensive and stochastic evaluation. An optimizer, an LLM paired with a coding harness, receives a target agent's seed harness, graded evaluation feedback, and a fixed target-evaluation budget. It edits the harness and nominates a final candidate, which is scored by its normalized gain over the seed on a held-out test partition that remains inaccessible throughout search. A trusted execution environment enforces the evaluation boundary, meters target-agent resource use, and preserves candidate versions for audit. We evaluate 5 frontier LLMs as optimizers both under a shared coding harness and under their native harnesses across 4 downstream tasks, over 111 scored runs. Experiment results show that optimizer models separate more than the coding harnesses they act through, native harnesses are not consistently superior, and gains vary substantially across tasks and seed regimes. These results establish harness optimization as a measurable and discriminative capability with large space for improvement.","upvotes":20,"discussionId":"6a753b60e1228e04b323814d","organization":{"_id":"6677220f8a4064c02bc81217","name":"ScaleAI","fullname":"Scale AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d6a5f94c28026a003581b4/uqHyTuNQ8fX7LheVhzPeO.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6a6a8246b172d8c070b522b7","avatarUrl":"/avatars/9c4be1d7e4c61864b187ec0e4565adc6.svg","isPro":false,"fullname":"Susan Jones","user":"susan-jones","type":"user"},{"_id":"6a6a9312c125cc860a9301b8","avatarUrl":"/avatars/6a252683ec681d412ebbb7bd73433ef6.svg","isPro":false,"fullname":"Thomas Sanchez","user":"thomassanchez","type":"user"},{"_id":"6a6aa101b7f4852961dd6caf","avatarUrl":"/avatars/dcd9edd4361eff66d69911dea531334c.svg","isPro":false,"fullname":"Patricia Davis","user":"Orbit-TrailV","type":"user"},{"_id":"6a6aa41d4c287dbb8e805dd3","avatarUrl":"/avatars/f109124053777f7667ff3235cd8aaf79.svg","isPro":false,"fullname":"Sarah Thompson","user":"rapidWing","type":"user"},{"_id":"6a6c7faa3139af1ea8b15fec","avatarUrl":"/avatars/0760a6faaefb1e99f1e39f7818fac226.svg","isPro":false,"fullname":"Joseph Martin","user":"cobaltflow","type":"user"},{"_id":"6a6c846fa3e5e6b7b047d92d","avatarUrl":"/avatars/1a3d200bdd027ce628bf7a7150c91e3c.svg","isPro":false,"fullname":"John Anderson","user":"Cedar-Kai","type":"user"},{"_id":"6a6c8532da65172f47ef3f9d","avatarUrl":"/avatars/4c207375dd1c09ba60f2cc666d6b01eb.svg","isPro":false,"fullname":"Brian Williams","user":"EmberGlade","type":"user"},{"_id":"6a6c8b9dd98e3eb6532ac650","avatarUrl":"/avatars/eb9a91990e1268fd9c43b6885104c63e.svg","isPro":false,"fullname":"John Williams","user":"atlasCraft","type":"user"},{"_id":"6a6d43cae1c088f52420afb3","avatarUrl":"/avatars/90870bb92e6c143ba173694c33629187.svg","isPro":false,"fullname":"Edward Williams","user":"Orbit-Noah","type":"user"},{"_id":"6a6c9c7aa1aab08eb34a1057","avatarUrl":"/avatars/edc8e84ec7eb19de792c37266fe48176.svg","isPro":false,"fullname":"Patricia Wilson","user":"patricia-wilson","type":"user"},{"_id":"6a6dc77f7ad403b19cd2e1fe","avatarUrl":"/avatars/9bf9b8366944d21c161cf7d2c9914b8e.svg","isPro":false,"fullname":"Mark Miller","user":"mark-miller","type":"user"},{"_id":"6a6de53ec51edbf08f1121ac","avatarUrl":"/avatars/8bcda53af64042ad674b08dd754ab666.svg","isPro":false,"fullname":"Karen Smith","user":"KarenSmith","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6677220f8a4064c02bc81217","name":"ScaleAI","fullname":"Scale AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d6a5f94c28026a003581b4/uqHyTuNQ8fX7LheVhzPeO.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.06301.md","query":{}}">
HarnessOpt-Bench: Evaluating LLMs at Harness Optimization
Abstract
As LLMs are increasingly deployed within agentic systems, their capabilities depend not only on the model weights but also on the harness: the prompts, tools, control flow, memory, and orchestration code surrounding them. This makes automated harness optimization -- the iterative and evaluation-guided improvement of a harness by an AI system -- both an important route to improving AI systems and a demanding capability for AI systems themselves. Yet the community lacks a common protocol for measuring how well frontier LLMs perform at this task. We introduce HarnessOpt-Bench, a benchmark for end-to-end harness optimization under expensive and stochastic evaluation. An optimizer, an LLM paired with a coding harness, receives a target agent's seed harness, graded evaluation feedback, and a fixed target-evaluation budget. It edits the harness and nominates a final candidate, which is scored by its normalized gain over the seed on a held-out test partition that remains inaccessible throughout search. A trusted execution environment enforces the evaluation boundary, meters target-agent resource use, and preserves candidate versions for audit. We evaluate 5 frontier LLMs as optimizers both under a shared coding harness and under their native harnesses across 4 downstream tasks, over 111 scored runs. Experiment results show that optimizer models separate more than the coding harnesses they act through, native harnesses are not consistently superior, and gains vary substantially across tasks and seed regimes. These results establish harness optimization as a measurable and discriminative capability with large space for improvement.
Community
This is an interesting direction for evaluating how well LLMs can actually improve agent systems, not just solve individual tasks. The focus on held-out evaluation and stochastic, budgeted optimization makes the benchmark especially useful for measuring genuine harness improvement rather than overfitting.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.06301 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.06301 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.06301 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.