Skip to content
View original post on X: Sophia YangX· Same story78/100AI score78/100

Reflection AI releases Beam, an open-weight coding model under Apache 2.0

AISummary

Reflection AI has released Beam with open weights under the Apache 2.0 license, along with weights, a tokenizer, a reference implementation, and the open-source Mirror CLI coding tool.

The model has 501B total parameters with 23B active, a 1M context window, and scores 81.6 on SWE-bench Verified and 80.1 on Terminal-Bench 2.1. The author reports a 1.9B drafter model that speeds up generation 2–3x.

Post on XView on X
Sophia YangVerified on X
@sophiamyang

TL;DR

- 501B total parameters with 23B active, built for agentic coding. Text only, 1M context window, Apache 2.0 license.

- Quality close to the top open models while using about 3x fewer reasoning tokens.

- Benchmarks: 81.6 on SWE-bench Verified, 80.1 on Terminal-Bench 2.1 and 90.6 on GPQA Diamond. Artificial Analysis ranks it #2 among US open-weight models.

- 4-bit holds up: the NVFP4 version scores about the same as full precision, for example 80.7 versus 81.6 on SWE-bench Verified.

- Drafter: this is quite interesting actually - they shiped a 1.9B drafter model to speed up generation. 2–3x faster even with 32 users at once, for example 288 to 874 tok/s on GSM8K.

- Coding tool: they also open-sourced Mirror CLI, the coding tool Beam was trained in.

Reflection@reflection_ai
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: https://huggingface.co/reflection
View quoted post on X

This story is featured in top picks: “Reflection releases Beam, a 501B open MoE model for coding and agents”

Source: Sophia Yang · x.comPublished

Topics

AI agentsAI codingOpen-source ecosystemModel releasesEvals & benchmarks