Poolside launches Laguna S 2.1, an 118B open-weight model built for agentic coding and long-horizon work, that it says competes with larger open models
ForbesIain Martin
Context & Ripple Effects
Poolside’s open-weight strategy has moved quickly from its earlier 33B-A3B Laguna XS.2 release to a substantially larger model aimed at more extended software tasks. That progression makes this a product-positioning step, not just a new checkpoint.
The launch also lands in an active contest over open models for coding and reasoning, following Z.ai’s GLM-5 open-weight push in coding and agentic tasks. Poolside is now making a direct performance claim within that category.
First-order effects
Developers and organizations evaluating self-hosted coding models gain another 118B-weight option for agentic and long-horizon workflows, with Poolside positioning it against larger open alternatives.
Poolside broadens its public model lineup and gives prospective users a concrete open-weight product on which to assess its coding-focused claims.
Second-order effects
Competing open-model providers targeting coding agents face added pressure to demonstrate task reliability and usable deployment trade-offs, rather than compete on parameter scale alone.
Model buyers can compare more suppliers for agentic coding deployments, strengthening their ability to choose models around workload fit and operational control.
Third-order effects
If more coding-model vendors publish capable weights, durable differentiation is likely to shift toward tooling, evaluation, integration, and support around the model rather than access to the weights alone.
The release reinforces a market split in which open-weight models become a credible route for organizations that want greater control over deployment, while performance claims increasingly need validation in real software workflows.
The trend: Agentic coding is becoming a key battleground for open-weight models, with vendors pairing larger releases with claims of practical long-horizon capability.
Poolside Laguna S2.1; a step up from before. 40+ on DeepSWE is a good number. But perhaps svg and html aren't where it shines. We shouldn't judge all model on same standard. [image]
I'm very curious about this one. Like I been running 2 x 3090s on Qwen 3.6 27b and 2 x 3090 on Qwen 3.6 35B A3.... A some point I got to move on right... Would this be solid to run on all 4 ? It can fit.
Poolside was an early AI star raising a $500 million round back in 2024 but then largely disappearing from view. Now it has launched a big new open-source AI model it claims can beat China's best https://www.forbes.com/...
52 days from kickoff to a 118B/8B MOE beating models 4-25x its size isn't a benchmark flex, it's a claim about the factory that made it. Publishing the trajectories instead of just the scorecard is the right instinct - lets go & find the holes.
the more i look into it, the more impressive it is! best in class:) notable things were fp8 training for RL and scaling RL with big focus on terminal use related tasks!! [image]
Today we're releasing Poolside Laguna S 2.1 It is a 118B-total, 8B-active open-weight model built for agentic coding and long-horizon work, with context up to 1M tokens https://poolside.ai/... Laguna S 2.1 sits at the top of its weight class and competes with open models many
Quite excited about this one: Laguna S 2.1: 118B-A8B MoE with 1M context. It runs on a single NVIDIA DGX Spark, and it stacks up very well on what we focus on: agentic coding. A few observations we had building it. 🧵
Today we are releasing Laguna S 2.1. At 118B total parameters, with 8B active per token, it does the work of models several times its size on agentic coding. It is remarkably persistent across long-horizon tasks. And it is small enough to run on a single NVIDIA DGX Spark. It is
Big day for American open-source AI. For the launch of Laguna S, I sat down with @eisokant to discuss its architecture, the economics of open weights, and the question of who gets to build intelligence. Timestamps: 0:00 Intro 1:50 Why Poolside started opening its models: the [vid…