Compare models
| Detail |  Laguna S 2.1 |  Laguna XS 2.1 |  Laguna M.1 |
|---|---|---|---|
| Best for | Long-horizon agentic coding and extended tool-use workflows | Fast agentic coding and rapid iteration | Complex, multi-step coding tasks where agent quality matters more than raw throughput |
| Parameters | 118B total, 8B active per token | 33B total, 3B active per token | 225B total, 23B active per token |
| Estimated total GPU memory (FP8) | 192 GB | 96 GB | 384 GB |
| Context window | 1M tokens | 256K tokens | 256K tokens |
| Model type | Mixture of Experts | Mixture of Experts | Mixture of Experts |
| Modality | Text-to-text | Text-to-text | Text-to-text |
| Reasoning | Native reasoning with thinking on or off per request | Native reasoning with thinking on or off per request | Native reasoning with thinking on or off per request |
| Model weights | Hugging Face | Hugging Face | Hugging Face |
| License | OpenMDW-1.1 | OpenMDW-1.1 | Apache 2.0 |
Reasoning terminology
Poolside and compatible providers use related terms for model reasoning behavior:- Reasoning: The general model capability or behavior. In API contexts, this can also refer to fields such as OpenRouter’s
reasoningobject or assistantreasoning_content. - Thinking: The Poolside term for turning reasoning behavior on or off per request.
- Effort: A generic scale used to express how much reasoning to request. ACP clients support
default,max,xhigh,high,medium,low,minimal, andnone; OpenRouterreasoning.effortvalues depend on provider model metadata. - Thought level: The ACP session config option that maps to effort values.
Get started
- Try a Poolside model in your browser. No sign-up required.
- Follow the quickstart to use a model with
pool. - Use the Poolside API for base URLs, model IDs, authentication, and request examples.