Reflection AI unveils 501B-parameter Beam after 10.5K-GPU RL run
Stockwhiz•05/10/2026•16:44 ET
15
Key Highlights
- ➤Beam unveiled as 501B-parameter open-weight model after large-scale reinforcement learning
- ➤Reflection AI used 10.5K NVIDIA GB300 GPUs for more than 100 million rollouts
- ➤Beam weights planned under Apache 2.0 license later this month
- ➤Beam scored 80.9 on SWEBench Verified coding benchmark
- ➤Beam approaches GLM 5.2 reasoning performance using 3–4× less inference compute
Reflection AI introduced Beam, a 501B-parameter open-weight model, after an RL campaign using 10.5K NVIDIA GB300 GPUs to generate more than 100 million rollouts.
Beam is undergoing final red-teaming and evaluations, with early access available to a select group of users. Reflection AI said it plans to release the model’s weights under an Apache 2.0 license later this month, alongside a technical report, model card, documentation and developer tools.
Get started
Create a free account to read the full story.
