Laguna M.1/XS.2 Technical Report: MoE Foundation Models for Agentic Coding

Forum topic · 小凯 · 2026-05-29

Summary

This arXiv paper (2605.27605) presents Laguna M.1 and Laguna XS.2, two Mixture-of-Experts foundation models built for long-horizon, agentic coding. Laguna M.1 has 225.8B total parameters with 23.4B activated per token, while XS.2 has 33.4B total parameters with 3B activated. Both models were trained from scratch end-to-end inside an internal system called the Model Factory—a tightly integrated stack of versioned data, training, evaluation, and inference components that turns model development into an industrial process. The paper describes the Model Factory's design principles, pre-training data and architecture, post-training stages, evaluation, and quantization. On agentic software engineering and terminal benchmarks including SWE-bench Verified, SWE-bench Multilingual, SWE-Bench Pro, and Terminal-Bench 2.0, both models are competitive with state-of-the-art open-weight models at their respective size levels. Laguna XS.2 weights are released under the Apache 2.0 license on Hugging Face.

Paper Overview

Field: AI Authors: Julien Abadji, Marah Abdin, Connor Adams, et al. Published: 2026-05-28 arXiv: 2605.27605

Highlights

Laguna M.1: a Mixture-of-Experts foundation model with 225.8B total parameters (23.4B activated per token), built for long-horizon, agentic coding.
Laguna XS.2: a smaller MoE model with 33.4B total parameters (3B activated per token).
Both models were trained from scratch end-to-end inside the same internal system, the Model Factory: a tightly-integrated stack of versioned data, training, evaluation, and inference components that turns model development into an industrial process.
The report details the Model Factory's principles and design choices, as well as the full end-to-end training process: pre-training data and architecture, post-training stages, evaluation, and quantization.
On agentic software engineering and terminal benchmarks — SWE-bench Verified, SWE-bench Multilingual, SWE-Bench Pro, and Terminal-Bench 2.0 — M.1 and XS.2 are competitive with state-of-the-art open-weight models at their respective sizes.
Laguna XS.2 weights are open-sourced under the Apache 2.0 license on Hugging Face.

Original Abstract (excerpt)

> We present Laguna M.1 and Laguna XS.2, two Mixture-of-Experts foundation models built for long-horizon, agentic coding: M.1 has 225.8B total parameters (23.4B activated per token) and XS.2 has 33.4B total (3B activated). Both models were trained from scratch end-to-end inside the same internal system that we refer to as our Model Factory: a tightly-integrated stack of versioned data, training, evaluation, and inference components that turn model development into an industrial process.

--- *Auto-collected on 2026-05-29*