Laguna XS.2

Overview

Poolside's first open-weight release: a 33B MoE (3B active) under Apache 2.0 built specifically for agentic coding. Scores 68.2% on SWE-Bench Verified and runs on a 36 GB Mac.

When to pick this model

• Local coding agents on developer laptops (Mac 36 GB or similar)
• Apache 2.0 commercial coding assistants
• Agentic workflows needing native tool calls and streaming
• Frontier-grade SWE-Bench scores without datacenter hardware
• Replacing closed coding APIs with a self-hosted alternative

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	19 GB
Q5_K_M	23 GB
Q8_0	35 GB
FP16 (no quantization)	66 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Published benchmark scores

Benchmark	Score
SWE-Bench Verified	68.2
SWE-Bench Multilingual	62.4
SWE-Bench Pro	44.5
Terminal-Bench 2.0	30.1

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

Strengths

• 68.2% SWE-Bench Verified — top-tier among open models
• Runs on a 36 GB Mac
• Apache 2.0 with no commercial restrictions
• Native tool calls and streaming
• Official Ollama tag with multiple quantizations

Limitations

• Coding-specialized — not a general chat model
• MoE + SWA architecture needs transformers v5.6.2 or newer
• Interleaved thinking can slow first-token latency

Architecture & training

Architecture: MoE 33B/3B active · 256 experts + 1 shared · 40 layers (10 global attention + 30 sliding-window 512) · FP8 KV-cache · 128k ctx

Training: Poolside's first open-weight model, optimized for local agentic coding. Muon optimizer, BF16, native reasoning with interleaved thinking.

Quick start

ollama run laguna-xs.2

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.