Laguna S 2.1 (118B-A8B, Q4)
Poolside's Laguna S 2.1, a 118B-parameter Mixture-of-Experts model with about 8B active parameters per token, built for agentic coding and long-horizon software work. It supports native tool calling and interleaved reasoning between tool calls. Choose the official imatrix Q4_K_M GGUF (about 75GB) for llama.cpp, or the calibrated oQ4e MLX build (about 64GB) for native Apple Silicon inference on a 96GB-or-larger Mac. Gezel defaults to Poolside's quality-recommended 256K context window; the underlying checkpoint can be extended to 1M context with explicit YaRN overrides.
At a glance
- Parameters
- 118B
- Quantization
- Q4_K_M
- Context window
- 262K tokens
- Approx. size
- 75.2 GB
- Engines
- llama.cpp, MLX
- License
- OpenMDW-1.1 (open)
- Version
- v1.0.0 , released 2026-07-21
- Family
- glm
Tuning
Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.
- sampling.temperature
0.7- sampling.topP
0.95- sampling.maxTokens
8192- reasoning.thinkingBudget
4096- reasoning.enableThinking
true- profiles
- thinking-general, thinking-coding, thinking-precise, instruct, creative
Model behaviors
Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).
- reasoning.strip-think-tags
- prompt.private-reasoning-guidance
- fabrication.detect-past-tense-no-tools
- turn.preamble-folding
- turn.ramble-detection
- tools.mlx-grammar
Sources
- llama.cpp (GGUF)
- poolside/Laguna-S-2.1-GGUF ·
laguna-s-2.1-Q4_K_M.gguf· Q4_K_M - MLX
- mlx-community/Laguna-S-2.1-oQ4e
- Upstream
- https://huggingface.co/poolside/Laguna-S-2.1
- Manifest
- View on GitHub
Tags
- poolside
- coding
- agentic
- tools
- reasoning
- long-context
- moe
- large
- open-weights
- apple-silicon