Model Overview & Capabilities
Beam-1.0-501B-A23B is Reflection AI’s premier open-weights reasoning architecture released on October 16, 2026, securing Rank #22 with an index score of 35.5. Utilizing 501 billion total parameters and 23 billion active parameters per token under an Apache 2.0 license, it achieves verified benchmark scores of 36.0% on SWE-bench Verified, 35.0% on Terminal-Bench 2.0, 36.0% on Aider Polyglot, and 74.0% on GPQA Diamond.
Hosted cloud endpoints offer Beam 1.0 at $0.25 per million input tokens and $1.00 per million output tokens, which yields an effective blended price of $0.44 per million tokens under 3:1 weighting and an economical $0.0123 per index point. Beam generates code at 125 tokens per second with a 1.15-second time to first token.
Featuring a 1,000,000-token context window and permissive open-source licensing, Beam 1.0 gives enterprise engineering teams complete freedom to self-host private code completion services, run continuous repository refactoring, develop custom agents, and deploy localized models without third-party data transmission risks. It establishes a trustworthy community-driven alternative for corporate environments requiring full architectural sovereignty.
