Reflection released Beam on October 6, 2026, saying it matches GLM 5.2 on key benchmarks while using three to four times less compute. It is the company's first open-weight model update since its founding in 2024.
Reflection reported Beam scores 44.4 on the Agentic Coding Benchmarks, measured against GLM 5.2 and Qwen 3.8. That compares with 61.0 on the same benchmark for GLM 5.2.
Beam is built on a mixture-of-experts architecture and targets coding, logical reasoning, and agentic tasks. Availability begins later this month, initially for select users.
"Beam activates 23 billion of its 501 billion parameters per token, keeping compute costs relatively low," said Jonathan Kemper, the source. The model is designed to balance efficiency with performance.
The announcement follows Reflection's $2 billion funding round in October 2025, positioning it as a Western counterpart to Deepseek and Qwen. Reflection said it is going after businesses that use AI for coding and automated workflows but need to keep operating costs in check.
Reflection did not say how it will close the remaining gap to top-performing open models, and noted that stronger models like Kimi K3 still beat it on raw performance. The company said it is already training a successor to address this.
Source: thedecoder
