Archive · 6 Oct 2026
Nvidia-Backed Reflection Unveils Beam to Challenge Chinese AI Models
Reflection AI has introduced an AI model called Beam.
It is the startup’s first model whose weights are open.
Reflection says Beam is meant to help with coding and tasks where AI agents take actions.
The company wants it to compete with lower-cost models from China.
Reflection says Beam performs comparably to one Chinese model on reasoning tests and is approaching another on coding and agent tasks.
Beam has 501 billion parameters, which are parts that help the model process information.
For each task, it activates 23 billion of them.
Reflection says this could make Beam faster and cheaper to run.
The company plans to share more technical details later this month.
Reflection AI introduced Beam, its first open-weight model, on Monday.
The startup says Beam is designed for coding and AI agent tasks, with business customers and developers in mind.
Reflection says Beam performs comparably to Z.AI’s GLM-5.2 on advanced reasoning benchmarks and is closing in on Qwen3.8-Max for coding and agentic tasks.
The company says Beam has 501 billion total parameters and activates 23 billion for each task.
Reflection says using part of the model can make it faster and cheaper to run; more technical details are expected later this month.
- Who
- Reflection AI, a startup backed by Nvidia.
- What
- It debuted Beam, its first open-weight AI model, aimed at coding and AI agent tasks.
- When
- Monday; the articles do not specify a date.
- Why
- Reflection aims to compete with lower-cost AI models, including Chinese open models, and serve business customers and developers.
- Model
- Beam, Reflection AI’s first open-weight model
- Launch
- Debuted on Monday; no date specified
- Total parameters
- 501 billion, according to Reflection
- Parameters activated per task
- 23 billion, according to Reflection
- Intended tasks
- Coding and AI agent tasks
- Performance comparisons
- Reflection says Beam is comparable to Z.AI’s GLM-5.2 on advanced reasoning benchmarks and is closing in on Qwen3.8-Max for coding and agentic tasks.
- Further details
- Reflection plans to release more technical details later this month.
