{"rewrite":{"id":"r_6c794e10768531b08fc41d95","clusterId":"c_012aff0f3bf560947c3dc9b9","slug":"reflection-unveils-beam-a-501b-open-weight-model-aimed-at-chinese-open-source-ai","model":"deepseek-v4-1-flash","headline":"Reflection Unveils Beam, A 501B Open-Weight Model Aimed At Chinese Open-Source AI","summary":"Reflection announced Beam, an open model with 501 billion parameters and 23 billion active parameters. It was pretrained on 23.8 trillion tokens and trained with reinforcement learning across 10,500 NVIDIA GB300s for four weeks, generating over 100 million rollouts. Reflection says Beam matches GLM 5.2 on coding and agentic tasks at one-third to one-quarter the inference compute, and approaches Qwen 3.8-Max.","whyItMatters":"Reflection is pitching Beam's inference efficiency, not raw capability, as the lever against larger Chinese open models, and the weights are promised for October 2026.","webCardHtml":"\u003cp\u003eBeam is not out yet. Reflection says the model is still in final-stage red teaming and evaluation, and that it plans to release the full weights in October 2026, with a waiting list open through its playground.\u003c/p\u003e\u003cp\u003eThe comparison chart puts Beam against GLM 5.2, Qwen 3.8-Max, Inkling, and Nemotron Ultra across DeepSWE, Terminal Bench, HLE No Tools, SWE Bench Pro, SWE Bench Verified, and CritPT AA. Reflection\u0026#39;s own claim is parity on those benchmarks, with the efficiency gap widening against models above 2 trillion parameters.\u003c/p\u003e","blueskyPost":"Reflection trained Beam on 10,500 NVIDIA GB300s for four weeks to claim compute efficiency. Efficiency claims are the argument, not the benchmark.","twitterPost":"Reflection says Beam matches GLM 5.2 at a third to a quarter of the inference compute. The efficiency claim is the pitch.","threadsPost":"Reflection trained Beam on 10,500 NVIDIA GB300s for four weeks to claim it matches GLM 5.2 at a third to a quarter of the inference compute. The efficiency number is the actual pitch. Matching quality by spending less per query is what open-weight buyers will check first.","newsletterBlurb":"Reflection announced Beam, a 501-billion-parameter open-weight model with 23 billion active parameters, pretrained on 23.8 trillion tokens and put through four weeks of reinforcement learning on 10,500 NVIDIA GB300s. Reflection says Beam matches GLM 5.2 on coding and agentic benchmarks while using one-third to one-quarter the inference compute, and approaches Qwen 3.8-Max. The model is still in final-stage red teaming, with full weights promised for October 2026.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261006-beam-reflection-501b-open-weight-model/\",\"title\":\"Reflection Unveils 501-Billion-Parameter Open Model \\\"Beam\\\" to Counter Chinese Open-Source AI, Claims Parity With GLM 5.2 at One-Third to One-Quarter the Compute and Parity With Qwen3.8-Max on AI Agent Tasks\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4641,"outputTokens":704,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1791270439,"createdAt":"2026-10-06T07:05:28.000Z","publishedAt":"2026-10-06T07:06:12.000Z","updatedAt":"2026-10-06T07:06:12.000Z"},"cluster":{"id":"c_012aff0f3bf560947c3dc9b9","canonicalTitle":"中国のオープンソースAIに対抗するためReflectionがパラメーター数5010億のオープンモデル「Beam」を発表、GLM 5.2に匹敵し使用する計算量は3～4分の1でAIエージェントタスクではQwen3.8-Maxに匹敵すると主張","representativeArticleId":"a_e3017ec94b96e04c750229e8","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"Beam\"],\"studios\":[\"Reflection\"],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-10-06T03:06:00.000Z","lastSeenAt":"2026-10-06T03:06:00.000Z","updatedAt":"2026-10-06T07:06:07.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261006-beam-reflection-501b-open-weight-model/","title":"中国のオープンソースAIに対抗するためReflectionがパラメーター数5010億のオープンモデル「Beam」を発表、GLM 5.2に匹敵し使用する計算量は3～4分の1でAIエージェントタスクではQwen3.8-Maxに匹敵すると主張"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["Beam"],"studios":["Reflection"],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":null}