MODEL · POOLSIDE · 33B TOTAL / ~3B ACTIVE (SPARSE MOE), TRAINED FROM SCRATCH ON 30T TOKENS
Laguna XS 2.1 (33B-A3B)
The first open-weight release from Poolside, a coding-first lab that trains from scratch rather than fine-tuning another base. A 33B-total / ~3B-active MoE aimed at agentic software engineering — the same big-but-sparse shape that made Qwen3-Coder-30B-A3B the community daily driver. Ships with official BF16 / FP8 / INT4 / NVFP4 checkpoints, an official GGUF repo, and DFlash speculator models that Poolside claims roughly double local decode speed.
License: OpenMDW-1.1 (Linux Foundation permissive model license — commercial OK, not classic-OSI) · Context: 256K (hosted API); long-context local support depends on runner · Released: July 2, 2026 (weights public with the announcement; HF repo staged June 20)
The decision in five lines
- The call
- Consider — runnable locally, family reference
- Best for
- Local evaluation and family reference
- Runs on
- 16 hardware picks fit (cheapest: Minisforum UM890 Pro · $463)
- Watch out
- General chat or non-coding work — it is a coding specialist.
- Evidence
- Estimated
- 33B total
- PARAMETERS
- MOE
- TYPE
- 256K
- CONTEXT
- ~20 GB (INT4) — fits a single 24 GB GPU
- VRAM AT Q4
Where we recommend this
This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.
The call
The first open-weight release from Poolside, a coding-first lab that trains from scratch rather than fine-tuning another base. A 33B-total / ~3B-active MoE aimed at agentic software engineering — the same big-but-sparse shape that made Qwen3-Coder-30B-A3B the community daily driver. Ships with official BF16 / FP8 / INT4 / NVFP4 checkpoints, an official GGUF repo, and DFlash speculator models that Poolside claims roughly double local decode speed.
When not to use: General chat or non-coding work — it is a coding specialist. All headline benchmarks are vendor-reported so far, so verify it against your own repo before displacing a proven pick like Qwen3-Coder-30B-A3B or North Mini Code. OpenMDW-1.1 is permissive, but read it if your org requires classic OSI licenses.
Runner notes
Runs under vLLM / SGLang / TensorRT-LLM / transformers; official GGUFs exist (llama.cpp support maturing — check current release notes). Try it free on OpenRouter (`poolside/laguna-xs-2.1`) before downloading ~20 GB. The official DFlash speculators are worth pairing for local decode speed on supported runners.
Hardware that fits
Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.
- Minisforum UM890 ProGood · 1.1× 32 GB DDR5 (shared) · $463–$580 all-in
- AMD Radeon RX 7900 XTXGood · 1.1× 24 GB · $810 used / ~$1,340 new
- NVIDIA RTX 3090 (used, single)Good · 1.1× 24 GB · $950–$1,200
- MacBook Air M5 24 GBRequires tweak · 1.0× 24 GB unified · $1,499–$1,899
- Mac Mini M4 Pro 24 GBRequires tweak · 1.0× 24 GB unified · $1,599
- Dual RTX 3090 (used)Perfect · 2.2× 48 GB · $1,800–$2,500 all-in
- NVIDIA RTX 4090Good · 1.1× 24 GB · $2,200–$2,800
- Framework Desktop (Ryzen AI Max+ 395)Perfect · 4.0× 128 GB unified · $2,459–$2,851 (128 GB config)
- M5 Pro MacBook Pro 48 GBPerfect · 1.5× 48 GB unified · $2,999–$3,599
- NVIDIA RTX 5090Perfect · 1.5× 32 GB · $3,500–$4,300
- NVIDIA RTX A6000 (48 GB, used)Perfect · 2.2× 48 GB ECC · $3,500–$4,500
- Mac Studio M4 Max 64 GBPerfect · 2.0× 64 GB unified · $3,799
- NVIDIA DGX SparkPerfect · 4.0× 128 GB unified · $4,699
- M5 Max MacBook Pro 64 GBPerfect · 2.0× 64 GB unified · ~$5,199 (est.; June 25 2026 increase)
- Mac Studio M3 Ultra 96 GBPerfect · 3.0× 96 GB unified · $5,299
- Dual RTX 5090Perfect · 3.0× 64 GB (2×32) · $8,500–$10,500
Next step
Find-by-model — see what hardware runs this→