--- title: Reflection launches Beam, an open model with 501B parameters description: Reflection AI released its first model, Beam, described as a highly efficient agentic open model with 501B total parameters and 23B active parameters. Early... canonical: https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Reflection launches Beam, an open model with 501B parameters | daily.dev og:description: Reflection AI released its first model, Beam, described as a highly efficient agentic open model with 501B total parameters and 23B active parameters. Early... og:url: https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf og:image: https://api.daily.dev/og/posts/lyI3HUcwF.png og:image:alt: Reflection launches Beam, an open model with 501B parameters 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. # Reflection launches Beam, an open model with 501B parameters **[Collections](https://daily.dev/sources/collections)** · 2 min read · 3 upvotes · 0 comments ## Summary Reflection AI released its first model, Beam, described as a highly efficient agentic open model with 501B total parameters and 23B active parameters. Early reactions compare it to GLM-5.2 in capability, with some calling it potentially the best American open-source model to date. Weights are expected to be released soon, so real-world benchmarking against GLM-5.2 is still pending. ## Content Reflection AI, the Nvidia-backed startup founded by former Google DeepMind researchers, has released its first model. It is called Beam, and it is meant for coding, reasoning and agentic work. The weights haven't shipped yet. Reflection says they will be released this month under an Apache 2.0 license, along with FP8 and NVFP4 builds. ## What Beam is Beam is a sparse mixture-of-experts model with 501B total parameters and 23B active. It is text-only. Pretraining covered 23.8T tokens and took under four weeks on 6,144 Nvidia GB300 GPUs. The effective context is 1M tokens. The reinforcement learning run used 10,500 GB300s for four weeks. According to Reflection, it produced more than 100M rollouts across nearly 1M environments and about 1.3B sandboxes, and it ended with no sign of a plateau. ## How it compares Reflection's pitch is efficiency, not raw capability. It says Beam is competitive with China's GLM-5.2 and approaching Qwen 3.8-Max, while using 3-4x less inference compute than GLM-5.2. Moonshot AI's Kimi K3 is still ahead on raw capability. The efficiency number needs a caveat. It rests on estimated forward-pass FLOPs (2 × active parameters × generated tokens). That leaves out prefill, attention and serving overhead, so it is not a measured cost. On Terminal Bench v2.1, Beam scores 80.1 against GLM-5.2's 81.0. It trails DeepSeek V4.1 Flash at 90.6 and Kimi K3 at 88.3. Neither of those appears in the headline chart. ## Safety review Two government safety bodies are reviewing the model. Reflection plans to open-source its internal safety tests alongside a technical report and model card. ## Reaction Most of the response has been enthusiastic. @iScienceLuvr called it possibly the best American open-source model and welcomed another strong US open-source player. @omarsar0 pointed to the reasoning efficiency as a big deal for long-running agents. @rohanpaul_ai called it the strongest Western open model. @AlexFinn was the most excited. He said America has entered the frontier open-weights race, and he compared Beam to GLM 5.2, Qwen 3.8 and Opus 4.8. He suggested a Mac Studio with 512GB would be enough to run it locally. That is his estimate, not a Reflection spec. @ClementDelangue offered a fair counterpoint: you can't be the best open-weight model if you aren't open-weight yet. The claims will be easier to judge once the weights are out and people can test them. ## Questions this post answers ### What are the parameter counts for Reflection AI's Beam model? Beam has 501 billion total parameters with 23 billion active parameters, using a mixture-of-experts style architecture for efficiency. It is described by Reflection AI as a highly efficient agentic open model. The weights have not yet been publicly released, so independent benchmarking against models like GLM-5.2 is still pending. _Developers tracking open-weight model releases can follow comparisons like Beam versus GLM-5.2 on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 18 comments across x (as of 2026-10-07). **TL;DR:** Replies are mostly skeptical banter, questioning claims of being the 'best open-weight model' before weights are even released, with some outright calling it overhyped compared to existing open models. **Sentiment:** 20% positive · 45% mixed · 35% skeptical **The case for** - A few replies acknowledge the model seems decent or at least appreciate that someone is trying to compete in open weights. **The pushback** - Many replies mock the idea of claiming 'best open-weight' status before weights are actually public and verifiable. - Some argue it's merely on par with smaller existing open models like Qwen, not a genuine leap forward. **By community** - x (skeptical): Replies lean toward mocking premature 'best open-weight' claims and doubting the model's real-world performance until weights are released. **Hottest debate:** Whether it's valid to call something the best open-weight model before the weights are actually public and independently verifiable. **Open questions** - How long is the typical gap between claiming a top benchmark and actually shipping weights? - When will the technical report be published? **Highlights** > @ClementDelangue "you can't be the best open-weight model if you're not open-weight yet!" 🚨 BASED ALERT 🚨 > — [bijanbowen on x · 26 points, 1 comments](https://x.com/bijanbowen/status/2107525801063383280) > @bijanbowen @ClementDelangue they can if they say so, also i call cap... they barely on par with qwen 3.8 27b > — [soyamiruku on x](https://x.com/soyamiruku/status/2107526792865321371) > @ClementDelangue Open weights also means anyone can rerun the evals. Until October 31, the best open-weight claim for Le Chonk is a number on a slide nobody outside Mistral can check. > — [DoDataThings on x](https://x.com/DoDataThings/status/2107536807885951295) > @ClementDelangue >best open weight model that isn’t from where people make good open weight models lol glad somebody’s trying, i guess > — [GingerDotDev on x](https://x.com/GingerDotDev/status/2107534562385678419) **Source threads** - [x](https://x.com/ClementDelangue/status/2107525319012090301) · 0 points · 18 comments ## Similar posts on daily.dev - [Together AI expands fine-tuning service with more models, live metrics, and finer controls](https://daily.dev/posts/together-ai-expands-fine-tuning-service-with-more-models-live-metrics-and-finer-controls-jagreco8n) · Together AI · 0 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts), [#glm](https://daily.dev/tags/glm) [View this post on daily.dev](https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf) ```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":"Reflection launches Beam, an open model with 501B parameters","url":"https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf"},"datePublished":"2026-10-05T19:50:21.314Z","dateModified":"2026-10-07T14:13:22.613Z","description":"Reflection AI released its first model, Beam, described as a highly efficient agentic open model with 501B total parameters and 23B active parameters. Early...","isAccessibleForFree":true,"articleSection":"Collections","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":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,llm,mixture-of-experts,glm","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Reflection launches Beam, an open model with 501B parameters"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/reflection-launches-beam-an-open-model-with-501b-parameters-lyi3hucwf#faq","mainEntity":[{"@type":"Question","name":"What are the parameter counts for Reflection AI's Beam model?","acceptedAnswer":{"@type":"Answer","text":"Beam has 501 billion total parameters with 23 billion active parameters, using a mixture-of-experts style architecture for efficiency. It is described by Reflection AI as a highly efficient agentic open model. The weights have not yet been publicly released, so independent benchmarking against models like GLM-5.2 is still pending. Developers tracking open-weight model releases can follow comparisons like Beam versus GLM-5.2 on daily.dev."}}]} ```