Reflection AI's Beam-1.0 tops US open-weight models, hitting 88% of Qwen3.8-2.4T with 23B active params
ryanshrout · x · 2026-10-27
Signal65 benchmarked Reflection AI's newly released Beam-1.0 (501B total, 23B active) on its PINNACLE agentic suite, calling it the best US open-weight model tested.
- Ranks fourth among 59 open-weight rows, behind only Alibaba's Qwen3.8 family, reaching 88% of the Qwen3.8-2.4T score with a quarter of the active parameters.
- The deployable 4-bit NVFP4 build scores at parity with FP8/BF16 repos and holds 192 agents at 10 tok/s on a single NVIDIA B300 node, completing 96.1% of multi-step enterprise jobs with a 1.5% fabrication rate.
- PINNACLE uses deterministic code-verified scoring with runtime-generated sandboxes and answer keys — no model judges, nothing static to memorize.
More from Infra
- Anthropic locked into $475B compute deals, OpenAI lifts projected spend to ~$750B through 2030 — kyliebytes · 2026-10-27
- First systematic data center noise study: 17 of 25 hyperscalers drew no credible complaints — AndyMasley · 2026-10-27
- Fork builds engine to run Qwen at ~400 tok/s on three used V100s — jfinkpottery · 2026-10-27
- RFD4-Proteina runs on an RTX 3060: protein design from $0.001 per design — AllThingsApx · 2026-10-27
- NPM packages now take ~20 minutes to go live, dev suspects AI-attack throttling — cocktailpeanut · 2026-10-27
- Nature Physics: reprogrammable refractive index enables arbitrary multimode wave control for ML — MarioKrenn6240 · 2026-10-27