The open-weights race just got an american entry. reflection's beam is a 501b-parameter model with…
the open-weights race just got an american entry. reflection's beam is a 501b-parameter model with 23b active, trained in 4 weeks on 10.5k nvidia gb300s, with apache 2.0 weights promised later this month, per oct 5 announcement.
it claims 4x less inference compute while still trailing kimi k3 on deepswe.
Context
Reflection's announcement says Beam is its first open-weight model: a sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active, built for coding, reasoning and agentic workloads, pretrained on 23.8 trillion tokens. It says the weights will be released under an Apache 2.0 license this month, with a technical report, model card and developer artifacts, and that the model is in final red-teaming and evaluations.
The Next Web reports Reflection said it pretrained Beam in under four weeks on 6,144 Nvidia GB300 GPUs, and that a reinforcement learning run then ran for four weeks on 10,500 GB300 GPUs. Reflection says Kimi K3 remains ahead on raw capability and that Beam's advantage is efficiency at inference time. Its table lists DeepSWE v1.1 scores of 44.4 for Beam and 68.0 for Kimi K3.
The note says Beam was trained in 4 weeks on 10.5k GB300s. The sources read split this in two: pretraining took under four weeks on 6,144 GPUs, and the four-week reinforcement learning run used 10,500 GPUs. The post joins the two figures.
The note says 4x less inference compute. The Next Web reports Reflection's claim as three to four times less inference compute than GLM 5.2 on advanced reasoning benchmarks, and says Reflection described it as an approximate comparison, not a measured cost. All benchmark numbers are Reflection's own and were not independently verified in the pages read.
The weights are promised, not yet released, in the announcement read, which matches the note's 'promised later this month'. 'American entry' is the author's framing. Reflection's own phrase is the Western open-weight frontier, and the Next Web headline calls it Nvidia-backed.
Related work
- Introducing Beam: Reflection's 501B open-weight model (Reflection, October 5, 2026) ↗Primary source for the parameters, tokens, license plan and benchmark table.
- Nvidia-backed Reflection unveils Beam, its first open-weight model (The Next Web, October 5, 2026) ↗Source for the GPU counts, the compute comparison and the vendor-figures caveat.
Watch next
- Check the technical report when it is released. Look for independent benchmark runs against Kimi K3.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 05:00 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →