{"type":"post","id":"ibm-research-compared-cuga-hermes-9603fb","slug":"ibm-research-compared-cuga-hermes-9603fb","title":"IBM Research compared CUGA, Hermes","description":"IBM Research compared CUGA, Hermes, and LangChain’s Deep Agents on the same open-weight model and 90 AppWorld tasks.","retrieval_nugget":"IBM Research compared CUGA, Hermes, and LangChain’s Deep Agents on the same open-weight model and 90 AppWorld tasks. Task completion ranged from 30% to 81% across harnesses, a useful reminder that the software around a model can shape results—but the vendor-authored, task-specific test is not a general ranking.","published_at":"2026-11-03","updated_at":"2026-11-03","record_date":"2026-11-03","date_kind":"published_at","topics":["hermes"],"entities":["IBM Research","CUGA","Hermes","Deep Agents","GLM-5.3-flash","AppWorld","Harbor","LangChain"],"related_patterns":["harness-architecture-outlives-model-choice"],"novelty_label":"new_fact","what_changed":"IBM reports its own CUGA-versus-Hermes-and-Deep-Agents test on 90 AppWorld tasks, with harness changes shifting completion from 30% to 81%, and emphasizes enforceable enterprise rules and deployment boundaries.","prior_refs":["https://newruntime.com/posts/a-smarter-model-won-t-save-your-bad-36f226/"],"source_url":"https://huggingface.co/blog/ibm-research/cuga-harness","source_urls":["https://huggingface.co/blog/ibm-research/cuga-harness"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:ibm-research-compared-cuga-hermes-9603fb","status":"published","visuals":[],"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/ibm-research-compared-cuga-hermes-9603fb/","markdown":"https://newruntime.com/posts/ibm-research-compared-cuga-hermes-9603fb.md","json":"https://newruntime.com/posts/ibm-research-compared-cuga-hermes-9603fb.json"},"markdown_url":"https://newruntime.com/posts/ibm-research-compared-cuga-hermes-9603fb.md","next_reads":[{"type":"topic","path":"/topics/hermes/","url":"https://newruntime.com/topics/hermes/","title":"Hermes","reason":"Explore the hermes topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/hermes.md","json_url":"https://newruntime.com/topics/hermes.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}