Laguna XS 2.1

Laguna XS 2.1 is a 33B-parameter Mixture-of-Experts model with 3B parameters activated per token, designed for agentic coding and long-horizon work on local machines.

Highlights

• Agentic coding: trained for software engineering, terminal tasks, and multi-step tool use.
• Efficient MoE architecture: routes each token through 8 of 256 experts plus one shared expert.
• Native reasoning: supports interleaved thinking between tool calls with a per-request thinking toggle.
• Long context: supports up to 262,144 tokens.