--- title: Holo4: powering generalist computer-use agents | daily.dev description: H Company released Holo4, a new generation of agentic models available in 27B dense and 35B-A3B mixture-of-experts sizes, plus an updated Holotron4 Nano built... canonical: https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Holo4: powering generalist computer-use agents | daily.dev og:description: H Company released Holo4, a new generation of agentic models available in 27B dense and 35B-A3B mixture-of-experts sizes, plus an updated Holotron4 Nano built... og:url: https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw og:image: https://api.daily.dev/og/posts/bj7NDJ4tW.png og:image:alt: Holo4: powering generalist computer-use agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Holo4: powering generalist computer-use agents **[Hugging Face](https://daily.dev/sources/huggingface)** · 8 min read · 0 upvotes · 0 comments ## Summary H Company released Holo4, a new generation of agentic models available in 27B dense and 35B-A3B mixture-of-experts sizes, plus an updated Holotron4 Nano built on Nemotron 3 Nano Omni. Holo4 works across GUIs, code, MCP, and APIs using a single model and calling convention across desktop, web, and Android. On OSWorld 2.0 it scores 61.7% (27B) and 30.9% (35B-A3B), trailing frontier closed models like Opus 5.5 (81.8%) but at a fraction of the cost. The models were trained using an internal Agentic Task Factory that has generated roughly 10,000 verifiable tasks, plus a rebuilt harness for longer-horizon agentic workflows. Weights are open on Hugging Face in multiple quantizations, and trajectories behind the benchmark scores are published for replay. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Holo4 27B compare to frontier closed models on OSWorld 2.0? Holo4 27B scores 61.7% on OSWorld 2.0, compared to 81.8% for Opus 5.5, while using orders of magnitude fewer parameters and costing much less per task. The 35B-A3B mixture-of-experts variant reaches 30.9%. Both models are built on a Qwen base and trained with H Company's Agentic Task Factory and reinforcement learning. _Developers evaluating agentic model trade-offs between cost and capability can track releases like Holo4 on daily.dev._ ### What interfaces does the Holo4 agentic model support for computer-use tasks? Holo4 interacts with software through GUIs, code execution, MCP, and APIs, using whichever interface fits the task, unlike most agentic models that are trained for only one interface. It runs identically on desktops, the web, Android, code sandboxes, and against business APIs, called the same way in every case. _Teams picking an agentic model for cross-platform automation can follow updates on tools like Holo4 via daily.dev._ ### What is Holotron4 Nano and how was it created? Holotron4 Nano is a generalist agentic model produced by applying H Company's post-training recipe to NVIDIA's Nemotron 3 Nano Omni model, as a follow-up to Holotron 3. It significantly improves over the base Nemotron model on GUI workflows and in environments exposing MCP, APIs, or coding sandboxes, showing the training recipe transfers across model sizes. _Engineers adapting foundation models into agentic systems can follow Nemotron and Holotron developments on daily.dev._ ## Similar posts on daily.dev - [Holo3.1: Fast & Local Computer Use Agents](https://daily.dev/posts/holo3-1-fast-local-computer-use-agents-balqgi7sz) · Hugging Face · 0 upvotes · 0 comments - [High Throughput Computer Use Agent](https://daily.dev/posts/high-throughput-computer-use-agent-btrteool5) · Hugging Face · 0 upvotes · 0 comments --- Tags: [#mcp](https://daily.dev/tags/mcp), [#agentic-ai](https://daily.dev/tags/agentic-ai), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Holo4: powering generalist computer-use agents","url":"https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw"},"datePublished":"2026-09-28T09:47:55.469Z","dateModified":"2026-09-28T09:48:23.191Z","description":"H Company released Holo4, a new generation of agentic models available in 27B dense and 35B-A3B mixture-of-experts sizes, plus an updated Holotron4 Nano built...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b3290da6ddf43f157bb955cc271e58d2?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b3290da6ddf43f157bb955cc271e58d2?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"mcp,agentic-ai,qwen","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"Holo4: powering generalist computer-use agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/holo4-powering-generalist-computer-use-agents-bj7ndj4tw#faq","mainEntity":[{"@type":"Question","name":"How does Holo4 27B compare to frontier closed models on OSWorld 2.0?","acceptedAnswer":{"@type":"Answer","text":"Holo4 27B scores 61.7% on OSWorld 2.0, compared to 81.8% for Opus 5.5, while using orders of magnitude fewer parameters and costing much less per task. The 35B-A3B mixture-of-experts variant reaches 30.9%. Both models are built on a Qwen base and trained with H Company's Agentic Task Factory and reinforcement learning. Developers evaluating agentic model trade-offs between cost and capability can track releases like Holo4 on daily.dev."}},{"@type":"Question","name":"What interfaces does the Holo4 agentic model support for computer-use tasks?","acceptedAnswer":{"@type":"Answer","text":"Holo4 interacts with software through GUIs, code execution, MCP, and APIs, using whichever interface fits the task, unlike most agentic models that are trained for only one interface. It runs identically on desktops, the web, Android, code sandboxes, and against business APIs, called the same way in every case. Teams picking an agentic model for cross-platform automation can follow updates on tools like Holo4 via daily.dev."}},{"@type":"Question","name":"What is Holotron4 Nano and how was it created?","acceptedAnswer":{"@type":"Answer","text":"Holotron4 Nano is a generalist agentic model produced by applying H Company's post-training recipe to NVIDIA's Nemotron 3 Nano Omni model, as a follow-up to Holotron 3. It significantly improves over the base Nemotron model on GUI workflows and in environments exposing MCP, APIs, or coding sandboxes, showing the training recipe transfers across model sizes. Engineers adapting foundation models into agentic systems can follow Nemotron and Holotron developments on daily.dev."}}]} ```