Reflection AI releases open-weight model Beam
What happened
On 6 October 2026, Reflection AI released Beam, its first frontier open-weight model. According to TechCrunch, the company says Beam matches leading Chinese open models on advanced reasoning benchmarks at substantially lower inference cost; Beam is a text MoE model with 501 billion parameters and 23 billion active parameters, pretrained on 23.8 trillion tokens with a 1 million token context window, and the company says it uses 3-4 times less inference compute than rivals. The Decoder reported the same day that Beam is Reflection's first freely available model, uses a MoE architecture activating 23 billion of 501 billion total parameters per token, and targets coding, logical reasoning and agentic tasks. Latent Space reported the same day that Beam is a text-only MoE model for coding, agents and scientific research, with 501B total and 23B active parameters, trained from scratch, with full Apache 2.0 weights expected this month. Reports agree on the model's parameter and activation figures, with no contradictions so far.
Written by AI from the reports · updated 10 d ago
Timeline
Follow the reports to see the event from each side.
- The DecoderReflection's Beam becomes the most capable open-weight model built outside China
Reflection has released Beam, its first open-weight model. The mixture-of-experts system activates just 23 billion of its 501 billion parameters per token and aims to match GLM 5.2 on coding and reasoning while using three to four times less compute. The strategy against Chinese open-weight rivals like Deepseek and Qwen is efficiency over raw performance.
- Latent SpaceSelected[AINews] Reflection Beam - 501B-A23B American Open Model
Reflection announced Beam, a text-only 501B-total / 23B-active MoE for coding, agentic and scientific work. It was trained from scratch, and full weights under Apache 2.0 are due this month.
- TechCrunch · AIReflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
Reflection is aiming Beam and future models at enterprises and sovereign nations. The pitch is to build “AI factories,” a product that would let institutions build their own customized, local AI system by training Reflection’s AI models on their own proprietary data.