The AI Bench
Poolside · OpenMDW 1.1

Laguna 2.1

Poolside’s coding-agent models: XS (33B-A3B) for 24–36 GB machines, S (118B) for 128 GB.

How much memory it needs

The model file, plus memory for the conversation, plus about 1 GB for the app. Longer conversations need more.

SizeVersionFileAt 8KAt 32KAt 128K
33B (3B active)Q4_K_M20.6 GB21.9 GB23.0 GB27.2 GB
Q8_035.6 GB37.0 GB38.0 GB42.2 GB
118B (8B active)Q4_K_M96.0 GB97.5 GB98.7 GB104 GB
Q8_0129 GB130 GB131 GB137 GB

Which computers run it

The best size for each machine at an 8K-token conversation. Speeds are estimates from memory bandwidth.

ComputerBest versionFitSpeed
Laptop or PC, no graphics card, 32 GB33B Q4_K_MJust fits15–30tok/sSmooth
RTX 4060 · 8 GB33B Q4_K_MSpills into system RAM18–35tok/sSmooth
RTX 3060 · 12 GB33B Q4_K_MSpills into system RAM25–45tok/sSmooth
RTX 5060 Ti · 16 GB33B Q4_K_MSpills into system RAM30–60tok/sFast
RTX 4090 · 24 GB33B Q4_K_MJust fits100–210tok/sFast
RTX 5090 · 32 GB33B Q4_K_MRuns great140–280tok/sFast
Mac with M4 · 16 GB33B Q4_K_MToo big—
Mac with M4 Pro · 48 GB33B Q4_K_MRuns great40–85tok/sFast
Mac with M5 Max · 128 GB118B Q4_K_MRuns well30–70tok/sFast
Ryzen AI Max+ 395 · 128 GB33B Q8_0Runs great25–50tok/sSmooth
DGX Spark · 128 GB118B Q4_K_MRuns well14–30tok/sSmooth
RTX PRO 6000 · 96 GB33B Q8_0Runs great110–220tok/sFast
Mac Studio with M5 Ultra · 256 GB118B Q8_0Runs great25–55tok/sFast
Mac Studio with M3 Ultra · 512 GB118B Q8_0Runs great19–40tok/sSmooth

What it is good and bad at

Good at

  • Coding agents
  • S has a 1M-token context

Not good at

  • The official S Q4 file is unusually large (96 GB); others are about 70 GB

Start it

Install Ollama, then run one of these in a terminal. LM Studio users can search the same name in the app.

ollama run laguna-xs-2.1:latest
ollama run laguna-s-2.1:latest