Reflection AI unveiled Beam on October 5, 2026, its first open-weight model. The company's blog post and TechCrunch both describe it, and TechCrunch notes Axios had reported a launch was close.
What Reflection says Beam is:
- A sparse mixture-of-experts model with 501 billion total parameters and 23 billion active.
- Pretrained on 23.8 trillion tokens, with a 1 million token context window, per TechCrunch.
- Built for coding, reasoning and agentic work, and text-only, TechCrunch reports.
- Trained with a high-compute reinforcement learning run of over 100 million rollouts on 10.5K Nvidia GB300 GPUs over four weeks.
The performance claims, all from Reflection:
- Competitive with larger open models such as GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks.
- On advanced reasoning benchmarks, scores comparable to GLM-5.2 while using 3 to 4 times less inference compute.
- Models such as Kimi K3 remain ahead on raw capability. Reflection says Beam's advantage is efficiency at inference time.
Not available yet: Reflection says Beam is in final red-teaming and evaluation. It will release the weights, a technical report and a model card later this month. Early access is by sign-up. TechCrunch says the performance claims have not been independently verified.
Background, per TechCrunch: Reflection was founded in 2024 by two former Google DeepMind researchers and has raised about $4.7 billion, with a last valuation of $25 billion pre-money.