Laguna M.1/XS.2 Technical Report

Poolside has released Laguna M.1 and XS.2, two Mixture-of-Experts foundation models designed for agentic coding tasks, with the smaller XS.2 model open-sourced under Apache 2.0. Both models achieve competitive performance on software engineering benchmarks while introducing a vertically-integrated 'Model Factory' approach to streamlined AI development.

Poolside's release of Laguna M.1 and XS.2 represents a significant architectural and operational advancement in foundation model development. The company has constructed what it calls a 'Model Factory'—a fully integrated pipeline spanning data versioning, training, evaluation, and inference—effectively industrializing what remains largely artisanal in most AI labs. This systems-level integration addresses a critical bottleneck in model development: the fragmentation between disparate tools and processes that slows iteration cycles and complicates reproducibility.

The technical specifications reveal thoughtful design choices for efficiency. M.1's 225.8B parameters activate only 23.4B per token, while XS.2 activates 3B from 33.4B total. This Mixture-of-Experts architecture permits scaling without proportional increases in inference costs—a practical consideration for deployment at scale. The emphasis on 'agentic coding' and performance on SWE-bench Verified and Terminal-Bench 2.0 positions these models specifically for autonomous software engineering tasks, an area of intense commercial interest.

The decision to open-source XS.2 under Apache 2.0 signals confidence in the model's capabilities while building community adoption. For developers and organizations, this offers a capable alternative to closed proprietary models without licensing restrictions. The technical report's detailed documentation of training processes, quantization approaches, and architectural decisions provides valuable transparency often absent from competing releases.

The competitive parity with state-of-the-art models across weight classes suggests Poolside's Model Factory approach achieves efficiency comparable to established players. Developers should monitor whether the promised industrial efficiency translates into faster release cycles and improved model quality over time.

• →Laguna XS.2 is now available open-source under Apache 2.0 with strong performance on software engineering benchmarks
• →Poolside's integrated 'Model Factory' system streamlines model development from data through inference and quantization
• →Both M.1 and XS.2 use Mixture-of-Experts architecture to reduce inference costs while maintaining competitive performance
• →The models are specifically optimized for agentic coding tasks rather than general-purpose applications
• →Technical transparency around training processes and architectural choices differentiates this release from competing foundation models