Poolside Releases Laguna S 2.1, Outperforming Deepseek V4 Pro
Poolside has released Laguna S 2.1, a new 118-billion parameter Mixture-of-Experts (MoE) model designed for agentic coding and long-horizon tasks. The model claims to be cheaper than Deepseek V4 Flash while delivering better performance than Deepseek V4 Pro. This release intensifies the competition in the LLM market by offering a highly cost-effective alternative for complex coding and reasoning tasks. It showcases the capabilities of independent research labs pushing the boundaries of AI efficiency and performance. Laguna S 2.1 activates 8 billion parameters per token and supports an extensive context window of up to 1 million tokens in both thinking and no-thinking modes. It also scored 70.2% on the Terminal-Bench 2.1 benchmark.
## BACKGROUND
Mixture-of-Experts (MoE) is an AI architecture that routes inputs to specific specialized subnetworks, allowing large models to run efficiently by only activating a fraction of their total parameters. "Neolabs" are emerging private research institutions founded by veteran AI researchers that focus on long-term breakthroughs rather than immediate commercial products. Deepseek is a prominent AI research company known for its highly cost-effective and high-performing open-source models.