Reflection AI’s Beam opens to early access, with open weights due later in October
- Reflection AI opened early access to Beam, a 501-billion-parameter open-weight model, on October 5.
- Weights, a technical report, and a model card follow later in October under an Apache 2.0 license.
- Reflection says Beam matches Z.ai’s GLM-5.2 on reasoning tests with 3–4× less inference compute.
Reflection AI launched Beam, its first open-weight model, in early access on October 5. The model is wrapping up red-teaming, and the weights, technical report, and model card follow later this month.
The weights ship under an Apache 2.0 license, Reflection says, with the full stack to run, evaluate, and fine-tune Beam.
GLM-5.2 parity comes with 3–4× less inference compute
Beam is a sparse mixture-of-experts model with 501 billion parameters and 23 billion active parameters per token. It was pre-trained on 23.8 trillion tokens, and its context length reached 1 million tokens during mid-training.
Reflection says Beam matches Z.ai’s GLM-5.2 on advanced reasoning tests while using “3–4× less inference compute.” Claims not verified independently.
GLM-5.2 has around 744 billion parameters, 40 billion active. Reflection puts Kimi K3 ahead in raw capability and pitches Beam’s advantage as inference efficiency.
Against Thinking Machines Lab’s Inkling, released in July, Reflection’s tables show Beam ahead on four coding benchmarks with scores for both models.
On SWE Bench Pro v2-Hard, Beam scored 77.2 to Inkling’s 56.9. Inkling is multimodal, while Beam is text-only.
10,500 Nvidia GB300 GPUs ran for four weeks of RL
Reflection’s RL run used 10,500 Nvidia GB300 GPUs for four weeks, produced more than 100 million rollouts, and utilized about 1.3 billion sandboxes for training and grading.
It’s one of the largest RL runs by any open lab, the company says. Inkling, by Reflection’s count, trained on 30 million rollouts.
Reflection was set up in 2024 by Misha Laskin and Ioannis Antonoglou, who are both former Google DeepMind researchers. It has raised about $4.7 billion from investors, including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, most recently at a $25 billion pre-money valuation.
A year ago, it raised $2 billion at an $8 billion valuation, Cryptopolitan reported. Its SpaceX deal alone runs up to $6.3 billion, at $150 million a month for Nvidia GB300 capacity at Colossus 2 near Memphis, Cryptopolitan reported in June.
Beam also fuels Reflection’s “AI factory” pitch, which lets institutions train its models on their own data and run them on their own compute. Shinsegae Group is trying out a sovereign edition in South Korea.
“They’re kind of like rocket ships,” Laskin has said of his models’ path to the frontier. “To build a big rocket ship, it takes time.”
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
FAQs
What is Beam and who built it?
Beam is Reflection AI's first open-weight model, a 501-billion-parameter mixture-of-experts system.
When will Beam's weights be released?
Later in October 2026, under an Apache 2.0 license, after final red-teaming.
How does Beam compare to rival models?
Reflection says Beam matches Z.ai's GLM-5.2 on reasoning tests while using 3–4× less inference compute.
Randa Moses
Randa Moses is an editor and reporter at Cryptopolitan covering tech, AI, robotics, crypto, scams, and hacks. She has worked in the crypto space since 2017. She held roles at Forward Protocol, AmaZix, and Cryptosomniac. Randa holds a degree in Electrical and Electronics Engineering from the University of Bradford.
