BREAKING
Laguna Runs on NVIDIA DGX Spark
0
$
price class
0
GB
unified memory
0
PFLOPS
FP4 AI perf
Local and Hosted Laguna Models
Laguna XS.2
Apache 2.0
●
33B MoE, ~3B active
●
Local agentic coding
●
Runs on single devices
Laguna M.1
API
●
225B MoE, ~23B active
●
256K context window
●
NVFP4 quantization
Laguna-XS-2.1 Throughput on Spark
Q4_K_M
117
Q8_0 GGUF
52
NVFP4 256K
42
Software Stack Still Maturing
Agentic Coding Moves Local
AI NEWS BLITZ
Poolside says its next Laguna coding model runs locally on NVIDIA's DGX Spark desktop.