# Reflection AI launches Beam-1.0 501B model for coding and reasoning

> The new sparse Mixture‑of‑Experts model has 501 billion parameters total, 23 billion active, and is offered via Fireworks’ on‑demand API.

Oossa · 2026-10-16 · https://oossa.com/en/reflection-ai-launches-beam-1-0-501b-model-for-coding-and-reasoning

Reflection AI released Beam-1.0-501B-A23B, a sparse Mixture‑of‑Experts model designed for coding, reasoning, and agentic tasks. The model has 501 billion parameters in total, with only 23 billion active at any time. It is available through Fireworks’ on‑demand deployment, which runs on dedicated GPUs without rate limits.

## The facts

- 501 billion total parameters, 23 billion active
- On‑demand API provided by Fireworks

## Why it matters

Developers can access a very large model for code‑related and reasoning work without the usual usage caps.

## Sources & references

1. [Reflection Beam-1.0-501B-A23B](https://fireworks.ai/models/fireworks/beam-1p0-501b-a23b) – Fireworks AI

Last updated: 2026-10-16