Reflection's Beam: one vendor post, no second source yet
2026-10-10 · published 21:52 UTC · research report
Reflection says Beam is a 501-billion-parameter open-weight model. Only one fetched source reports it, so Signal Scout treats it as unconfirmed.
Beam, as Reflection describes it
Beam is the one model in this report that passed both qualification gates. A Reflection blog post dated October 5, 2026 introduces it as Reflection's first open-weight model. [s1]
The post describes Beam as a sparse Mixture-of-Experts model with 501 billion total parameters, 23 billion of them active. It says the model is built for coding, reasoning, and agentic workloads. [s1]
Why Beam is labelled unconfirmed
Beam has a source count of 1, which makes it noise/unconfirmed under Signal Scout's scoring. No second fetched source reports the same signal, so every detail above comes from the vendor's own post. [s1]
The retained text of that post does not state a licence or link to the weights. The "open-weight" label therefore rests on Reflection's own wording and has not been checked against a model card or an independent report. [s1]
A second name that could not be ranked
A second page carries the title "Aleph Alpha Kolibri: What Germany's 'Sovereign' Open-Weight AI Model Means for You". Only that title was retained. It gives no publication date, licence or release detail. [s2]
Because the date could not be verified, Kolibri is excluded from the ranking. It is also a single, title-only source, so it should be read as unconfirmed. [s2]
What this report does not cover
Only s1 qualifies as a usable source, against a target of at least three qualifying pages from the last 7 days. With no model at two or more sources, the report cannot yet separate signal from noise. [s1] [s2]
No Reddit or Hacker News thread was fetched, so community reaction to Beam or Kolibri is not covered here. Treat this piece as a single-source early look, not a confirmed trend. [s1] [s2]