Reflection AI Releases First Open-Weights Model Beam
10/05/2026 — 10/07, 00:17·8 sources (all)·9 reports (all)·In the 2026-10-07 briefing·In the 2026-10-06 briefing
Key takeaways
- Nvidia-backed Reflection AI has announced its first open-weights MoE model, Beam, featuring 501B total parameters and a 1M token context win…
Background & analysis
On October 4, Axios reported that Reflection, an AI startup backed by an $800 million investment from Nvidia, was preparing to release its first open-weight model. The model was expected to trail the top US frontier models at launch, but to compete directly with leading Chinese open-weight models and to help companies build proprietary AI systems at lower cost. A day later, Techmeme relayed the same reporting, noting that several Western open-weight models were set to arrive during the month, with Reflection's first model among them.
On Monday, October 6, Reflection AI officially unveiled Beam, its first open-weight model. According to Semafor, the startup — which bills itself as America's answer to open-source Chinese AI — says Beam excels at coding and agentic tasks, uses 3x-4x less compute than comparable models, and comes close to Qwen3.8-Max. Reflection says Beam rivals Z.ai's GLM-5.2 and is approaching Qwen3.8-Max on coding and agentic work. TechCrunch reported that Beam is a text-only MoE model with 501B total parameters and 23B activated, pretrained on 23.8 trillion tokens with a 1M-token context window, with weights due this month. The company claims parity with GLM-5.2 on advanced reasoning benchmarks at far lower token cost and inference compute, though those performance claims have not been independently verified. IT之家 put the total at 501 billion parameters with 23 billion activated per task, and compared it with GLM-5.2's roughly 744 billion total and 40 billion active parameters. DEV Community added that the weights are planned for release this month under Apache 2.0, and that Beam focuses on compute-heavy reinforcement learning for reasoning, coding, and agentic tasks.
Artificial Analysis said it had been given access by Reflection and was independently benchmarking Beam, with early indicators suggesting it may be one of the most token-efficient open models for its level of intelligence. SiliconANGLE noted the launch came months after Reflection raised funding at a $25 billion valuation, and reported a $6.3 billion deal with SpaceX to rent Nvidia GB300 NVL72 appliances. ETtech said Beam targets coding and agentic tasks and aims to compete with lower-cost Chinese models such as DeepSeek and Kimi. The weights have not yet shipped and third-party evaluation has only just begun; accounts differ on which model Beam matches, with some reports pointing to Qwen3.8-Max and others to GLM-5.2.
AI-generated from 9 reports · updated 46 days ago
Latest turnReflection AI, a two-year-old startup that had never shipped a public model, has unveiled Beam: a text-only 501B-parameter MoE that activates 23B parameters per token, pretrained on 23.8 trillion tokens with a 1-million-token context window and high-compute RL for reasoning, coding and agentic tasks. The company claims parity with Z.ai's GLM-5.2 on reasoning benchmarks at roughly a third to a quarter of the inference compute, with Apache 2.0 weights promised this month.
Related AI tools
- Heat index
- 25
- All sources
- 8
- All reports
- 9
- First seen
- 48 days ago
Reports on this story headlines open the original
Reflection AI, a two-year-old startup that had never shipped a public model, has unveiled Beam: a text-only 501B-parameter MoE that activates 23B parameters per token, pretrained on 23.8 trillion tokens with a 1-million-token context window and high-compute RL for reasoning, coding and agentic tasks. The company claims parity with Z.ai's GLM-5.2 on reasoning benchmarks at roughly a third to a quarter of the inference compute, with Apache 2.0 weights promised this month.
DEV Community · AIAI score 62
Nvidia-backed Reflection AI has released Beam, its first open-weight model, aimed at coding and agentic tasks. The startup is positioning it against lower-cost Chinese models such as DeepSeek and Kimi.
ETtechAI score 58According to SiliconANGLE, Reflection AI introduced Beam, an open-source large language model with 501 billion parameters. The release comes just months after the startup raised funding at a $25 billion valuation, and the report says it also signed a $6.3 billion deal with SpaceX to rent Nvidia GB300 NVL72 systems.
SiliconANGLEAI score 60Reflection AI, a startup backed by Nvidia, released Beam, its first open-weights model, with 501B total parameters and 23B activated per task under a sparse architecture. The company says Beam matches Z.ai's GLM-5.2 and is closing in on Qwen3.8-Max on coding and agent tasks.
IT之家AI score 60Artificial Analysis says Reflection has given it access to independently benchmark Beam, the lab's new open model. Early indicators suggest Beam will be one of the most token-efficient open models seen at its level of intelligence.
X:Artificial Analysis (@artificialanlys)First-partyAI score 41Nvidia-backed Reflection AI has unveiled Beam, its first frontier open-weight model, claiming it matches leading Chinese open models on advanced reasoning benchmarks at a fraction of the token cost and inference compute. Beam is a text-only MoE model with 501B total and 23B active parameters, pretrained on 23.8T tokens with a 1M-token context window; weights are due this month and the performance claims are not independently verified.
TechCrunch · AIAI score 60According to Semafor, startup Reflection AI has released its first model, an open-weight system called Beam. The company says Beam excels at coding and agentic tasks while using 3x-4x less compute than comparable models, and comes close to Qwen3.8-Max.
TechmemeAI score 65
Reports indicate that Nvidia-backed startup Reflection AI plans to launch its first open-weight model this month. The upcoming release aims to compete with top open-weight models from Chinese developers and has drawn significant industry attention.
TechmemeAI score 59Reflection, an AI startup backed by an $800 million investment from Nvidia, is preparing to release its first open-weight model. The model is expected to compete directly with leading open-weight models from China and help enterprises build proprietary AI systems at a lower cost.
华尔街见闻AI score 69
Other stories people are talking about
- Heat index 32OpenDesign launches OpenMotion plugin for agent-made video motion graphics1 sources
- Heat index 114Ex-DeepSeek Researcher Zhang Bo Open-Sources Local Agent Linger with 2k GitHub Stars3 sources
- Heat index 77Microsoft rolls out Copilot integrations and unveils MAI models at Ignite 20262 sources
- Heat index 61DeepSeek closes a 100-billion-yuan funding round at a valuation of about 615.5 billion yuan2 sources
- Heat index 1134RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

