--- title: s1: Simple Test-Time Scaling - Can 1k Samples Rival... description: Stanford researchers introduce S1, a simple test-time scaling approach that fine-tunes Qwen 2.5 32B Instruct on just 1,000 carefully curated samples (s1k... canonical: https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview? | daily.dev og:description: Stanford researchers introduce S1, a simple test-time scaling approach that fine-tunes Qwen 2.5 32B Instruct on just 1,000 carefully curated samples (s1k... og:url: https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc og:image: https://api.daily.dev/og/posts/2DDdW89Sc.png og:image:alt: s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview? 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. # s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview? **[AI Papers Academy](https://daily.dev/sources/aipapersacademy)** · 8 min read · 0 upvotes · 0 comments ## Summary Stanford researchers introduce S1, a simple test-time scaling approach that fine-tunes Qwen 2.5 32B Instruct on just 1,000 carefully curated samples (s1k dataset) to achieve strong reasoning performance rivaling OpenAI's o1-preview. The s1k dataset is built by filtering 59,000 samples down to 1,000 through quality, difficulty, and diversity stages, with reasoning traces generated via Google Gemini Flash Thinking. The key innovation is 'budget forcing' — a test-time scaling technique that controls how many thinking tokens the model generates by either forcing early exit or appending 'wait' tokens to extend reasoning. Results show S1 is dramatically more sample-efficient than alternatives like R1-distill, which requires 800x more samples, and demonstrates a clear accuracy improvement as test-time compute increases, though excessive forcing can cause repetitive loops. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Inference Scaling \(Test-Time Compute\): Why Reasoning Models Raise Your Compute Bill](https://daily.dev/posts/inference-scaling-test-time-compute-why-reasoning-models-raise-your-compute-bill-yujmco4f2) · Towards Data Science · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc) ```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":"s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview?","url":"https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc"},"datePublished":"2026-04-19T08:04:50.095Z","dateModified":"2026-04-30T02:18:11.115Z","description":"Stanford researchers introduce S1, a simple test-time scaling approach that fine-tunes Qwen 2.5 32B Instruct on just 1,000 carefully curated samples (s1k...","image":"https://i.ytimg.com/vi/S4QP4AR-C-Y/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/S4QP4AR-C-Y/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Papers Academy","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":"AI Papers Academy","logo":"https://media.daily.dev/image/upload/s--O3RGOCh9--/f_auto,q_auto/v1776585874/logos/aipapersacademy","url":"https://daily.dev/sources/aipapersacademy"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/s1-simple-test-time-scaling---can-1k-samples-rival-o1-preview--2dddw89sc","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,qwen","timeRequired":"PT8M","video":{"@type":"VideoObject","name":"s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview?","description":"Stanford researchers introduce S1, a simple test-time scaling approach that fine-tunes Qwen 2.5 32B Instruct on just 1,000 carefully curated samples (s1k...","thumbnailUrl":"https://i.ytimg.com/vi/S4QP4AR-C-Y/sddefault.jpg","uploadDate":"2026-04-19T08:04:50.095Z","duration":"PT8M","url":"https://api.daily.dev/r/2DDdW89Sc","embedUrl":"https://www.youtube.com/embed/S4QP4AR-C-Y"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Papers Academy","item":"https://daily.dev/sources/aipapersacademy"},{"@type":"ListItem","position":3,"name":"s1: Simple Test-Time Scaling - Can 1k Samples Rival o1-Preview?"}]} ```