[AINews] Reflection Beam - 501B-A23B American Open Model

Key points

• But they stayed “stealth” longer than Thinking Machines and it was not clear we would ever get a model launch out of them, as the broader open model ecosystem did not slow down one bigt for them.1
• Still, since this is US-trained from scratch, there’s a segment of the market that has been eagerly waiting for more options here, and more importantly, Reflection has now announced its arrival as a functional neolab!1
• Beam launch: Reflection announced Beam, a text-only 501B-total / 23B-active MoE for coding, agentic and scientific work.1
• Independent and critical reads: Artificial Analysis has early access and expects Beam to be among the most token-efficient open models for its intelligence (AA).1

Sources (2)

• [1][AINews] Reflection Beam - 501B-A23B American Open Model
  Latent Space · Oct 6, 06:28 AM

  It’s been over a year since Reflection launched with us with big goals on coding (and hinted about their RL approach):

  But they stayed “stealth” longer than Thinking Machines and it was not clear we would ever get a model launch out of them, as the broader open model ecosystem did not slow down one bigt for them.

• [2]AI: Off the Charts
  AI Supremacy · Oct 7, 11:46 AM

Extractive summary: sentences quoted from the sources.