Reflection AI's Beam takes on China's open models

What happened: Reflection AI unveiled Beam, a 501 billion parameter mixture of experts open model with 23 billion parameters active per token, trained on 23.8 trillion tokens. Early access opened Monday. Weights are due later in October under the Apache 2.0 license after red teaming. Reflection was recently valued at $25 billion.
The claim: Reflection says Beam beats GLM 5.2 on some tasks using a third to a quarter of the hardware, approaches Qwen 3.8 Max, still trails frontier models like Anthropic's Claude Fable 5.1, and is the first open model from a US startup to match the best Chinese models.
Austin's take: Austin calls it the strongest US bid yet in a space China has dominated. He argues open models level the playing field, worries about lock-in from Chinese open models, and notes a model like this could potentially run slowly on a MacBook.
Why it matters: Open weight AI has been China's game. A competitive US open model gives developers a non-Chinese option they can run themselves.
What to do: Wait for the weights and independent benchmarks before building on it.
Not confirmed yet: The performance claims are Reflection's own benchmarks, and you can't download the model yet. It matters only if the weights actually ship.
Sources: Full stories and sources

The Loop