Reflection AI debuts open-source Beam model with 501B parameters

The launch follows Reflection AI's funding round at a $25 billion valuation a few months earlier. Around the same time, the startup reportedly signed a $6.3 billion deal with SpaceX Corp. to rent Nvidia GB300 NVL72 appliances, systems that each contain 72 graphics cards. Reflection AI used those systems to train Beam, according to the company.

Reflection AI benchmarked Beam against GLM-5.2, an open-source LLM with about 250 billion more parameters. The company determined that Beam can perform some tasks better while using between one-third and one-fourth the hardware. Reflection AI also said Beam approaches the performance of Qwen 3.8-Max, a model with more than 2 trillion parameters. Many advanced open-source LLMs, including Qwen 3.8-Max and GLM-5.2, were created by Chinese companies. Beam is the first open-source model from a U.S. startup to show comparable or better performance, according to the company. Free LLMs still trail frontier models such as Anthropic PBC's Claude Fable 5.1.

Reflection AI began Beam's development by training a relatively small prototype model, then created a series of successively larger and more capable algorithms. That workflow eventually produced Beam Base, the foundation for Beam. The company developed Beam Base using a cluster of 6,144 graphics cards. It trained the model on 23.8 trillion tokens sourced from the public web and commercial sources. According to Reflection AI, the dataset included a significant amount of software code. The company created custom filters for each programming language to remove low-quality files.

Reflection AI developed Beam Base in under four weeks. It then performed midtraining, an optimization process that extended the model's context window and enhanced its reasoning capabilities. That phase set the stage for the third, most hardware-intensive phase of the training workflow.

Reflection AI spun up 10,000 GB300 graphics cards and used them to launch 1.3 billion reinforcement learning sandboxes. Those virtual environments are where an AI model learns new skills. Beam's sandboxes were optimized for tasks such as generating code, searching the web and running AI agents. Reflection AI said the reinforcement learning phase took only four weeks. The company avoided unnecessary delays by developing software that prevented malfunctions from interrupting the workflow. Its cluster achieved a median recovery time of eight minutes across the 71 errors that occurred during training.

Beam is initially available through an early access program. Reflection AI plans to release its weights, documentation and fine-tuning tools later this month.

Editor's Summary Reflection AI has released Beam, a 501-billion-parameter open-source model trained on Nvidia GB300 systems. The company says Beam can outperform larger open-source rivals on some tasks with less hardware, though free models still trail frontier systems. Reflection AI plans to publish weights and fine-tuning tools later this month.