Beam 1.0 501B A23B
Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows.
Architectural Profile & Capabilities
Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic workflows. It uses a sparse mixture-of-experts architecture with 501B total and 23B active parameters per token, and a... On ARMES, Beam 1.0 501B A23B operates via strict Zero Data Retention inference pipelines. User inputs, prompts, and completions are processed ephemerally and are never stored, logged, or indexed for model training.
Autoregressive dense transformer with grouped-query attention (GQA) and rotary position embeddings calibrated for low TTFT.
Recommended Workloads & Primary Use Cases
- •Deep reasoning passes utilize test-time compute which may increase time-to-first-token (TTFT).
- •Long-context retrieval beyond 200K tokens requires structured needle-in-haystack prompt framing.
- •Tool definitions should enforce strict JSON Schema validation constraints.
Calls to Beam 1.0 501B A23B are routed through strict Zero Data Retention inference channels. Prompts and outputs are never stored, indexed, or monitored by Reflection.