Laguna XS 2.1
33.4B parameter open-weight model. Sparse mixture-of-experts coding model — 33.4B total parameters across 256 experts with 8 active per token, so it reasons like a large model while activating a fraction of it. The Ollama library tag ships Q4_K_M at ~20 GB, which fits a single 24 GB card. Measured on an RTX 3080 (10 GB) it loaded fully into VRAM at ~8.8 GB resident and sustained ~40-43 tok/s, so a 12-16 GB card is a realistic floor rather than the 24 GB the file size implies. Requires Ollama 0.32.3 or newer — older clients reject the manifest outright with an HTTP 412. Emits visible reasoning before answering; treat its output as a draft to review, since it will answer confidently about repositories and APIs it has no knowledge of.
Poolside · Laguna
Model overview
Laguna XS 2.1 is Poolside's compact coding agent, the smaller sibling to Laguna S. It is a mixture-of-experts design: 33.4B parameters in total, but only 8 of its 256 experts run per token, which is why it delivers large-model coding behaviour at a size that fits consumer hardware. It supports tool calling and exposes an explicit reasoning mode, both of which matter if you are wiring it into an agent loop rather than a chat window. The 256K context window is genuinely large enough for repository-scale work. It is released under OpenMDW-1.1 rather than a standard OSI licence, so check the terms before commercial deployment.
Editorial review
VRAM figures are empirical estimates. Actual usage varies by runtime, context length, and system configuration. Verify on your specific hardware before production use.
Will Laguna XS 2.1 run on your machine?
Laguna XS 2.1 is 33.4B parameters and needs 21.5 GB of VRAM at Q4_K_M — 20 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.
VRAM by quantization
| Quantization | Weights | Needs (with overhead) | Quality |
|---|---|---|---|
| Q4_K_M | 20 GB | 21.5 GB | good |
Fit on common hardware at Q4_K_M
| Hardware | Memory the model can use | System RAM | Verdict |
|---|---|---|---|
| CPU Only | None (CPU only) | 16 GB | Too large |
| RTX 4060 Laptop | 8 GB | 16 GB | Too large |
| RTX 3060 (12GB) | 12 GB | 32 GB | CPU offload |
| RTX 4060 Ti (16GB) | 16 GB | 32 GB | CPU offload |
| RTX 3090 | 24 GB | 64 GB | Comfortable |
| Apple Silicon (Unified Memory) 36 GB | 27 GB of 36 GB | 36 GB | Comfortable |
| RTX 5090 | 32 GB | 64 GB | Comfortable |
Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.
Apple Silicon shares one pool of memory between the system and the GPU, so a model cannot use all of it. These rows apply the same 75% usable fraction the Compatibility Checker uses, which is why a 36 GB Mac is graded on less than 36 GB.
Ready to run this model locally?
Find a compatible interface in our Local AI Tools directory →