Reflection launches Beam, a large open AI model with cheaper compute

Reflection, a well-funded US AI startup, released Beam, a massive open-weight AI model it claims rivals China’s best open models in reasoning but costs far less to run. Beam is a 501-billion-parameter, text-only model trained using reinforcement learning to be highly efficient. Reflection’s tests say Beam matches top Chinese models on advanced benchmarks while using three to four times less computing power.

• Beam has 501 billion parameters, 23 billion active
• Trained on 23.8 trillion tokens; 1 million token context window
• Reflection claims Beam is 3-4x cheaper to run than rivals
• Company raised $4.7 billion and signed $7B in chip deals
• Aimed at enterprises, public sector, and developers