Reflection AI previews Beam, a 501B open-weight model, with weights due this month

Why it matters

【Key points】

Reflection AI introduced Beam on October 5, 2026 as "Reflection's first open-weight model": a sparse Mixture-of-Experts with 501 billion total parameters and 23 billion active, for coding, reasoning and agentic workloads. The blog says Beam "is undergoing final red-teaming and evaluations" and points to a waitlist, promising that "this month, we will release the weights under an Apache 2.0 license".

【Verifiable】

Primary: https://reflection.ai/blog/introducing-beam ; waitlist at https://platform.reflection.ai/ . Its verified X account posted the same claims that day.

As stated: pretraining on 23.8 trillion tokens in under four weeks on 6,144 NVIDIA GB300 NVL72 GPUs; RL ran on 10.5K GB300 GPUs for four weeks, generating more than 100 million rollouts.

Reported Beam scores include SWEBench Verified 80.9, Terminal Bench v2.1 80.1 and HLE without tools 36.2; on the same Terminal Bench table the blog lists Kimi K3 at 88.3 and DeepSeek V4.1 Flash at 90.6. Reflection calls Beam "competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max", says Kimi K3 "remains ahead on raw capability", and claims parity with GLM-5.2 on reasoning "while using 3-4x less inference compute". Rival scores come from Artificial Analysis and DataCurve.

【Our read】

Western open weights are coming back, but the yardstick has changed: Reflection's comparison set is Chinese open models, and on most raw-capability rows Beam sits below them. The pitch is intelligence per token, not the top score.

Until the weights ship this is a preview: no model card, no technical report, rival numbers from aggregators, nothing independently reproducible today.

The unusual disclosure is the RL scale, not the leaderboard; if the technical report lands with it, that beats another benchmark table for teams running high-compute RL.

Watch whether Apache 2.0 weights arrive this month, and whether the promised safety evaluations come with them.