Reflection AI unveils open AI model Beam — Fortune

US company Reflection AI, based in New York and backed by Nvidia, has unveiled its first open artificial intelligence model, Beam. The company positions it as a tool for enterprises, the public sector, and developers.

As Fortune reports, in its own benchmark tests, Reflection AI's Beam model showed results similar to GLM-5.2 from Beijing-based company Z.ai. That model was released in July and was replaced by the newer GLM-5.3 version in August.

Claims about the model's efficiency

Reflection AI said that Beam is three to four times more efficient than competing open models from Western companies in terms of token and computing costs during inference. According to the developer, the model was created using reinforcement learning with substantial computing resources, which made it possible to achieve competitive results in coding and agent tasks.

More current news is available on the UA.News Telegram channel Telegram.

Users can download open models, run them on their own hardware or cloud infrastructure, and refine them. Fortune notes that large companies are increasingly choosing such models because of control over data security and costs, although in many cases they lag behind closed models in performance.

Contracts for computing resources

Reflection AI was founded in March 2024 by former Google DeepMind researchers Misha Laskin, the company's chief executive officer, and Ioannis Antonoglou, the startup's president and chief technology officer. In May, the company entered into a partnership to supply its models to the US Department of Energy and the US Department of War.

According to CNBC, in June Reflection AI signed a $6.3 billion agreement with SpaceX to access Nvidia GB300 chips at the Colossus 2 data center. Bloomberg reported that in July the company also entered into a $1 billion agreement for computing capacity through cloud provider Nebius. Reflection AI said it is already training the next, larger model in its lineup.