Poolside

Laguna XS 2.1

Name: Laguna XS 2.1
Rating: 82 (90 reviews)
Author: Poolside

Frontera

HuggingFace

22.5KDescargas170Me gustaJun 2026Publicado262K tokensContextoOpenMDW 1.1Licencia75 FuerteCalidad

Laguna XS 2.1 (33.400001525878906B parameters) requires approximately 22.5 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 3B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 26 GB of VRAM.

Comenzar

— copia y pega para ejecutar en local

Copy-paste commands to run Laguna XS 2.1 on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "poolside/Laguna-XS-2.1" \
  --hf-file "Laguna-XS-2.1-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters33.4B (3B active)

Architecturemoe (MoE)

Context262K tokens

Modalitytext

Min RAM13 GB

Rec. RAM20.4 GB (Q4_K_M)

LicenseOpenMDW 1.1

FamilyLaguna

✓ Code✓ Reasoning

About this model

Laguna XS 2.1 is the compact member of Poolside's Laguna 2.1 family: a 33B-total Mixture-of-Experts model activating roughly 3B parameters per token. It shares the family recipe — 256 routed experts plus one shared expert, grouped-query attention and interleaved full/sliding-window layers — in a 40-layer stack sized for single-GPU agentic coding at 262K context.

•33B total / ~3B active — runs comfortably on a single 24GB GPU at Q4.
•Same router and attention recipe as Laguna S 2.1, at a quarter of the size.
•262K context with 1:3 global-to-sliding-window attention (window 512).
•Targeted at agentic coding and long-horizon tool use.

Modelos relacionados

Selecciones rápidas

Mejor económicoA

Mac mini M4 64GB~$1,099 — 14 tok/s

Mejor en generalS

RTX 5090 32GB~$1,999 — 144 tok/s

Mejor hardware

Mejores opciones para Laguna XS 2.1

RTX 5090 32GBS

32 GB

RTX PRO 4500 Blackwell 32GBS

32 GB

AMD Instinct MI100 32GBS

32 GB

NVIDIA V100 32GBS

32 GB

AMD Instinct MI60 32GBS

32 GB

Ejecutar este modelo

Laguna XS 2.1 on RTX 5090 32GB Laguna XS 2.1 on RTX PRO 4500 Blackwell 32GB Laguna XS 2.1 on AMD Instinct MI100 32GB

Opciones de cuantización

Estimaciones de VRAM por nivel de cuantización

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	4.8 GB	Very Low	—
Q2_0_G128	1.71	8.9 GB	Low	—
Q2_K	2	13.0 GB	Low	—
Q3_K_S	3	16.4 GB	Low	—
NVFP4	4	18.7 GB	Medium	—
Q4_K_M	4	20.4 GB	Medium	—
Q5_K_M	5	24.0 GB	High	—
Q6_K	6	27.4 GB	High	—
Q8_0	8	35.7 GB	Very High	—
F16	16	68.5 GB	Maximum	—

Quality benchmarks

Laguna XS 2.1 benchmark scores

Benchmark verified

Coding

SWE-bench Verified70.9%

HumanEval+—

Aider Polyglot—

LiveCodeBench—

Source: official · 2026-06-20

Compatibilidad de hardware

Estimaciones de encaje en todo el hardware

Abrir calculadora

Computing compatibility...

Desglose de memoria

Reference: RTX 2060 6GB

Weights20.4 GB

KV Cache0.6 GB

Runtime0.9 GB

Headroom0.6 GB

Preguntas frecuentes

FAQ — Laguna XS 2.1

Ver también

Guía de cuantización Metodología de puntuación Abrir calculadora