An open-weight 501B-parameter sparse mixture-of-experts AI model from Reflection, built for coding, reasoning, and agentic workloads.
Introducing Beam announcement
1 / 2Overview
What is Beam?
Highly efficient agentic 501B parameter open model
Beam is Reflection's first open-weight AI model: a sparse mixture-of-experts with 501 billion total parameters and 23 billion active per token, built for coding, reasoning, and agentic workloads.
The weights are published free under the Apache 2.0 license on Hugging Face in BF16, FP8, and NVFP4 formats, so teams can run a frontier-scale model on their own hardware, infrastructure, or cloud and customize it to their use cases. It ships with an open-source Rust-based harness for best performance.
Beam is also available through Reflection's official API, currently in public beta and free to start, plus a Mirror CLI for terminal access.
Platforms and languages
Beam Availability
Platforms
Languages
Capabilities
Beam Key Features
501B sparse mixture-of-experts
501 billion total parameters with 23 billion active per token, designed for efficient frontier-scale inference.
Apache 2.0 open weights
Weights published free on Hugging Face in BF16, FP8, and NVFP4 precision formats, plus a speculative decoding drafter model.
Built for coding and agents
Focused on coding, reasoning, and long-horizon agentic workloads, with strong results on Terminal-Bench, HLE, and DeepSWE benchmarks.
Open-source Rust harness
Ships with a dedicated open-source Rust-based harness tuned to get the best performance from the model.
Official API in public beta
First-party API at platform.reflection.ai is in public beta and free to start, alongside the Mirror CLI and self-hosted weights.
Best for
Who uses Beam?
Teams wanting a frontier-scale model on their own hardware, infrastructure, or cloud
Run and customize an open-weight agentic model for coding and reasoning workloads
Plans and access
Beam Pricing
Free
Weights free under Apache 2.0; official API in public beta and free to start; no public pricing published yet.
Common questions
Beam FAQs
What is Beam?
Beam is Reflection's first open-weight model: a 501B-parameter sparse mixture-of-experts (23B active per token) focused on coding, reasoning, and agentic use cases.
Is Beam free to use?
Yes. The weights are published free under the Apache 2.0 license on Hugging Face, and the official API is in public beta and free to start.
How can I access Beam?
Three ways: download the open weights from Hugging Face (BF16, FP8, or NVFP4), use the official API at platform.reflection.ai, or use the Mirror CLI.
What license are the weights released under?
Apache 2.0, which permits commercial use, modification, and distribution.
What is Beam best at?
Coding and agentic workloads. Reflection reports strong benchmark results on Terminal-Bench 2.1, HLE, and DeepSWE, and ships an open-source Rust harness for best performance.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
