--- title: Open-sourcing AstaBrief, the fast report-generation... description: Ai2 has open-sourced AstaBrief 8B, a small language model built on Qwen3-8B that generates cited scientific reports from a research question and retrieved... canonical: https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Open-sourcing AstaBrief, the fast report-generation model in Asta | daily.dev og:description: Ai2 has open-sourced AstaBrief 8B, a small language model built on Qwen3-8B that generates cited scientific reports from a research question and retrieved... og:url: https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu og:image: https://api.daily.dev/og/posts/uTwMnRfLu.png og:image:alt: Open-sourcing AstaBrief, the fast report-generation model in Asta 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. # Open-sourcing AstaBrief, the fast report-generation model in Asta **[Hugging Face](https://daily.dev/sources/huggingface)** · 14 min read · 1 upvotes · 0 comments ## Summary Ai2 has open-sourced AstaBrief 8B, a small language model built on Qwen3-8B that generates cited scientific reports from a research question and retrieved literature excerpts in a single pass. Trained with supervised fine-tuning and direct preference optimization on real Asta user queries rather than reinforcement learning, it powers Asta's new Fast mode, averaging 51.1 seconds per report versus 178.5 seconds for the Claude-powered Thinking mode (about 3.5x faster). Development focused heavily on data filtering—particularly citation density—to close the gap with proprietary models on answer precision and citation grounding. Early usage data shows 29.1% of 374 trial users returned for two-plus days, and the model, training data, and an example local-report workflow are released on Hugging Face and GitHub. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is AstaBrief 8B and how was it trained? AstaBrief 8B is an open-weights model, based on Qwen3-8B, that generates cited scientific reports from a research question and retrieved literature excerpts in a single pass. It was trained using supervised fine-tuning on 47K filtered real research queries plus direct preference optimization on about 6K judged preference pairs, rather than reinforcement learning, to keep the training recipe simpler and more stable. _Developers building retrieval-augmented or scientific LLM pipelines can track model releases like this one on daily.dev._ ### How much faster is Asta's Fast mode than Thinking mode for generating research reports? Fast mode, powered by AstaBrief 8B, averages 51.1 seconds per report compared with 178.5 seconds for Claude-powered Thinking mode, roughly 3.5 times faster across the full Asta pipeline. This speedup comes from generating the full report in one pass instead of summarizing and clustering snippets section by section. _Teams weighing speed versus quality tradeoffs in LLM report generation can follow comparisons like this on daily.dev._ ### Which data filtering signal most improved citation grounding in a fine-tuned report-generation model? Filtering out synthetic training reports with low citation density produced the strongest quality gains, more than combining multiple filters or sweeping learning rates. Citation density measured the share of statements with at least one supporting citation; low-density reports tended to contain large stretches of unsupported text, so removing them improved answer precision and citation quality in the resulting model. _Anyone fine-tuning models for grounded, citation-heavy output can dig into data-filtering lessons like these on daily.dev._ ## Similar posts on daily.dev - [How V7 gives AI agents institutional memory](https://daily.dev/posts/how-v7-gives-ai-agents-institutional-memory-qelllmnld) · OpenAI · 8 upvotes · 0 comments - [GPT-6 Astra: The AI That Doesn’t Just Answer. It Gets the Job Done](https://daily.dev/posts/gpt-6-astra-the-ai-that-doesn-t-just-answer-it-gets-the-job-done-zotuejbng) · Medium · 0 upvotes · 0 comments - [OpenAI will sell you Astra, but not the system that scored 98.6% on ARC-AGI-3](https://daily.dev/posts/openai-will-sell-you-astra-but-not-the-system-that-scored-98-6-on-arc-agi-3-mgkuhrp8j) · The New Stack · 1 upvotes · 1 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#deep-learning](https://daily.dev/tags/deep-learning), [#rag](https://daily.dev/tags/rag), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu) ```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":"Open-sourcing AstaBrief, the fast report-generation model in Asta","url":"https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu"},"datePublished":"2026-10-02T15:24:25.498Z","dateModified":"2026-10-02T15:28:14.113Z","description":"Ai2 has open-sourced AstaBrief 8B, a small language model built on Qwen3-8B that generates cited scientific reports from a research question and retrieved...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8385a4f4ff911ce3497e5416f8fcb5f6?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8385a4f4ff911ce3497e5416f8fcb5f6?_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/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,deep-learning,rag,qwen","timeRequired":"PT14M"} {"@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":"Open-sourcing AstaBrief, the fast report-generation model in Asta"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/open-sourcing-astabrief-the-fast-report-generation-model-in-asta-utwmnrflu#faq","mainEntity":[{"@type":"Question","name":"What is AstaBrief 8B and how was it trained?","acceptedAnswer":{"@type":"Answer","text":"AstaBrief 8B is an open-weights model, based on Qwen3-8B, that generates cited scientific reports from a research question and retrieved literature excerpts in a single pass. It was trained using supervised fine-tuning on 47K filtered real research queries plus direct preference optimization on about 6K judged preference pairs, rather than reinforcement learning, to keep the training recipe simpler and more stable. Developers building retrieval-augmented or scientific LLM pipelines can track model releases like this one on daily.dev."}},{"@type":"Question","name":"How much faster is Asta's Fast mode than Thinking mode for generating research reports?","acceptedAnswer":{"@type":"Answer","text":"Fast mode, powered by AstaBrief 8B, averages 51.1 seconds per report compared with 178.5 seconds for Claude-powered Thinking mode, roughly 3.5 times faster across the full Asta pipeline. This speedup comes from generating the full report in one pass instead of summarizing and clustering snippets section by section. Teams weighing speed versus quality tradeoffs in LLM report generation can follow comparisons like this on daily.dev."}},{"@type":"Question","name":"Which data filtering signal most improved citation grounding in a fine-tuned report-generation model?","acceptedAnswer":{"@type":"Answer","text":"Filtering out synthetic training reports with low citation density produced the strongest quality gains, more than combining multiple filters or sweeping learning rates. Citation density measured the share of statements with at least one supporting citation; low-density reports tended to contain large stretches of unsupported text, so removing them improved answer precision and citation quality in the resulting model. Anyone fine-tuning models for grounded, citation-heavy output can dig into data-filtering lessons like these on daily.dev."}}]} ```