Reflection just launched Beam which is a 501B open weight model for coding, reasoning and agents that it says can match GLM-5.2 while using ~4x less inference compute. The more interesting part is RL used 10,500 $NVDA GB300s versus 6,144 for pretraining and generated 100M+ rollouts showing agent training is becoming a huge inference workload. That helps explain why Reflection locked in $7B+ of compute across $SPCX and $NBIS through 2029 while positioning Beam as a U.S. open model for enterprises and sovereign AI.

Reflection joins the list of Nvidia & Thinking Machines who have released their strongest models and come up behind Chinese counterparts. There's a lot of ways people will overthink this, but the clearest takeaway should be that the Chinese are very very good at building LLMs.

Reflection AI, an artificial intelligence startup from two former Google DeepMind researchers, has unveiled a new open-weight model that it says rivals leading options in the US and China

Reflection, backed by Nvidia, is building an open-weight model aimed at both Chinese upstarts and the US frontier labs. > Businesses can download and customize it with private data starting this month, built as a domestic alternative to DeepSeek and Qwen > It can reportedly rival the most expensive closed systems in certain cases, though it's expected to trail the best overall > The company has committed over $7B to compute through 2029, including a $1B Nebius deal and $150M monthly payments to SpaceX's Colossus cluster since July Officials in Washington have already been briefed, timed with a wave of other open-weight releases from US labs this month Nvidia backing a lab building open models it could end up competing with its own customers over is the part worth watching.

Reflection rewards = an improved staking model You still get rewarded for holding on to the token, but the reward cycle is much faster, and the rewards itself are (often) more valuable + lasting. Probably one of the greatest innovations this cycle to incentivize bagholding.