BestLLMfor Your hardware. Your LLM. Your call.
The Local Copilot Kit APIOpen data Find my LLM
Model fiche

Laguna S 2.1

By Poolside · United States

Updated 2026-08-28

chat code reasoning moe
Parameters
118B
License
OpenMDW 1.1
Context
256k
VRAM (Q4)
68 GB
Released
13 July 2026

Overview

Poolside's Laguna S 2.1 is a 118B-parameter MoE model (8B active) built for agentic coding and multi-step reasoning, with a 256K context window and a permissive OpenMDW 1.1 license.

When to pick this model

  • Running autonomous coding agents that plan and execute multi-step refactors
  • Large-scale multi-file refactoring jobs that need the full 256K context
  • Commercial coding-agent deployments that require an open license
  • Complex reasoning tasks tied to code, like debugging or architecture decisions

VRAM requirements by quantization

VRAM REQUIRED (GB)24324880128Q4_K_M68 GBQ5_K_M84 GBQ8_0126 GBFP16236 GB
QuantizationVRAM required
Q4_K_M (recommended)68 GB
Q5_K_M84 GB
Q8_0126 GB
FP16 (no quantization)236 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.

In practice, Laguna S 2.1 is server-class even at Q4_K_M (68 GB). Stepping up to Q8_0 nearly doubles the footprint to 126 GB, and unquantized FP16 weights take 236 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Laguna S 2.1 needs roughly 153 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 32 tokens/sec on entry-level GPUs, on the order of 50 tokens/sec on a mid-range card, and up to 75 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches Laguna S 2.1 to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memoryExample cardsBest fit for Laguna S 2.1
8 GBRTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GBDoes not fit — needs 68 GB at Q4_K_M
12 GBRTX 5070, RTX 5070 Ti Laptop, RTX 4080 LaptopDoes not fit — needs 68 GB at Q4_K_M
16 GBRTX 5080, RTX 4080 Super, Radeon RX 9070 XTDoes not fit — needs 68 GB at Q4_K_M
24 GBRTX 4090, Radeon RX 7900 XTX, RTX 5090 LaptopDoes not fit — needs 68 GB at Q4_K_M
32 GBRTX 5090Does not fit — needs 68 GB at Q4_K_M

Which GPU should you buy to run Laguna S 2.1?

To run Laguna S 2.1 locally at Q4, you need ~68 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).

Check Apple Mac Studio price on Amazon →Check Apple Mac Studio price on Newegg →

As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.

Strengths

  • Strong agentic coding and reasoning performance
  • 256K context handles large codebases in a single pass
  • OpenMDW 1.1 is a genuinely open, commercially usable license
  • MoE efficiency: only 8B active params despite 118B total capacity

Limitations

  • ~68GB VRAM at Q4 puts it out of reach for most single-GPU setups
  • Narrow focus on coding — not the best pick for general-purpose chat

Typical workloads

In our catalog grid, Laguna S 2.1 is filed under Agentic Coding, Reasoning, Multi-File Refactor — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); multi-step reasoning and math-flavoured tasks.

The 256k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. It ships under the OpenMDW 1.1 license — commercial use is generally possible but read the specific terms before embedding it in a product.

Architecture & training

Architecture: MoE 118B / 8B active · 256K context

Training: Poolside, Laguna line optimized for agentic coding. Native reasoning, open OpenMDW 1.1 license.

Verdict

A serious open-license option for agentic coding at scale, if you have the VRAM to run it.

Quick start

ollama run laguna-s-2.1

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

Similar models worth comparing

Frequently asked questions

How much VRAM does Laguna S 2.1 need?

At the recommended Q4_K_M quantization, Laguna S 2.1 needs about 68 GB of VRAM. Q8_0 takes 126 GB, and unquantized FP16 weights take 236 GB.

Can Laguna S 2.1 run without a GPU?

Yes — with roughly 153 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does Laguna S 2.1 support?

Laguna S 2.1 supports a 256k-token context window (262,144 tokens).

Can I use Laguna S 2.1 commercially?

Laguna S 2.1 ships under the OpenMDW 1.1 license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.

How fast is Laguna S 2.1 on consumer hardware?

Our compatibility engine estimates on the order of 50 tokens/sec on a mid-range GPU and up to 75 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Laguna S 2.1 should I download first?

Start with Q4_K_M (68 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.

Tools

Is Laguna S 2.1 the right pick for you?

Compute self-hosted ROI → Back to catalog