Reflection Beam debuts: 501B-param MoE with 23B active scores 30 on AA Intelligence Index
ArtificialAnlys · x · 2026-10-16
Artificial Analysis published full benchmark details for Reflection Beam, a new open-weights reasoning model.
Specs and headline results
- Apache 2.0 license, weights on Hugging Face, released October 2026
- 501B total parameters with only 23B active per token, 1M-token context window
- Scores 30 on the AA Intelligence Index (median for comparable models: 19), ranking #19 of 120 overall
- Notably concise: generated 79M output tokens vs. a 130M median across the Index
Standouts and weak spots across the ten evaluations
- AutomationBench-AA: 48.5%; CritPt: 13.1% — more than double K2 Horizon (4.6%) and Qwen3.8 27B (5.4%)
- Weakest on Terminal-Bench 4.0 (3.0%) and AA-Briefcase (976 Elo)
- Speed and pricing data not yet available
More from Models
- Gemini 4 Argon now available in AI Studio for Google AI Pro subscribers — CurieuxExplorer · 2026-10-17
- Generative UI 101: Building Live Interfaces with Agents on Opus 5.5 and Rivals — EXM7777 · 2026-10-17
- OpenAI's new $500 tier and $200 plan throttling are driving power users away — cybrstg · 2026-10-17
- DeepSeek shows rapid vision gains: agent self-checks screenshots to build 3D scene — cedric_chee · 2026-10-17
- Full FP8 n-gram table visibly improves local Qwen3.8-Flash-Next at zero speed cost — sillynoobhorse · 2026-10-17
- Altman says OpenAI shelved a ready-to-launch model as too risky, warning of AI 'escape' scenarios — Aiden_Tech_Ai · 2026-10-17