Written by Jakub Rusinowski · Last updated August 15, 2026
Model library → Poolside Laguna XS 2.1 → Laguna XS 2.1 33B-A3B
33B total parameters with only 3B activated per token, which is what lets a model scoring 70.9% on SWE-Bench Verified sit on a single 24GB card. Mixed sliding-window + global attention (3:1 across 40 layers), sigmoid per-head gating, FP8 KV cache, 256K context. Released under OpenMDW-1.1 — the Linux Foundation / NVIDIA permissive model-weights framework — not Apache 2.0, so check the terms if you are redistributing.
Laguna XS 2.1 33B-A3B needs about 21 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
| Parameters | 33 Billion (3B active) |
| Context window | 262,144 |
| Architecture | Mixture-of-Experts (sliding-window + global attention, FP8 KV cache) |
| Provider | Poolside |
| Licence | OpenMDW-1.1 |
| Specified at | Q4_K_M |
| System RAM | 32 GB |
| Record updated | 2026-08-15 |
OpenMDW-1.1 — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.
Modelled on a reference NVIDIA RTX 4090 (24 GB), with no KV cache (this record has no published architecture). Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.
| Quant | Weights | VRAM needed | Est. speed | Fit on 24 GB |
|---|---|---|---|---|
| Q2_K | 10.8 GB | 11.6 GB | ~210 tok/s (est.) | Fits comfortably |
| Q3_K_M | 14.1 GB | 14.9 GB | ~194 tok/s (est.) | Fits comfortably |
| Q4_K_M | 19.9 GB | 20.7 GB | ~171 tok/s (est.) | Fits comfortably |
| Q5_K_M | 23.4 GB | 24.2 GB | ~24 tok/s (est.) | Offloads to system RAM (slow) |
| Q6_K | 27.1 GB | 27.9 GB | ~22 tok/s (est.) | Offloads to system RAM (slow) |
| Q8_0 | 35.1 GB | 35.9 GB | ~20 tok/s (est.) | Offloads to system RAM (slow) |
| F16 | 66.0 GB | 66.8 GB | — | Won't fit |
Want the memory numbers alone, at every quantization level and your own context length? Use the Laguna XS 2.1 33B-A3B VRAM calculator.
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
The cheapest catalogued GPU that runs Laguna XS 2.1 33B-A3B is the AMD Radeon RX 7900 XTX (24 GB).
Install Ollama, then run:
ollama run poolside-lagunaWeights on Hugging Face: poolside/Laguna-XS-2.1.
Quality scores as published by the model's authors or an independent evaluator — not throughput, and not measured by us.
| Benchmark | Score | Provenance |
|---|---|---|
| SWE-Bench Verified | 70.9 / 100 % | reported · https://poolside.ai/blog/introducing-laguna-xs-2-1 |
| SWE-bench Multilingual | 63.1 / 100 % | reported · https://poolside.ai/blog/introducing-laguna-xs-2-1 |
Best for: agentic coding, software engineering, repo level, consumer gpu, long documents.
← All Poolside Laguna XS 2.1 models | VRAM calculator | Check your own hardware