Poolside: Laguna XS 2.1

Laguna XS 2.1 is a text model from Poolside, listed as the latest coding agent model in the 33B-A3B category. It is a step forward from Poolside's earlier Laguna XS.2 model, which was released in April 2026, and the provider release date for this version is 2 July 2026. If you are already working with Poolside's agent-oriented models, this is the current iteration in that line.

The model takes text as input and returns text as output, so it fits the standard chat and completion pattern rather than multimodal work. Its context window is 262,144 tokens, which is large enough to hold substantial source files, long logs or extended conversation history in a single request. The maximum output is 32,768 tokens, so responses can be long, but they are still bounded, which matters when you are generating whole files or lengthy reasoning traces. As a coding agent model, it is designed around tool use and multi-step work rather than one-shot answers.

Because it supports tools and tool choice, practical uses lean toward agentic coding workflows: running a loop where the model inspects a repository, calls functions you expose, and edits or explains code across several turns. The large context helps when the task depends on wide surrounding code, and the output ceiling supports generating complete modules or patches in one go. Reasoning support means you can surface intermediate thinking when you want it, or leave it out of the user-facing response. General text tasks are possible too, but the clearest fit is code assistance and tool-driven automation.

You call it through api14's OpenAI-compatible API using the model id "poolside/laguna-xs-2.1". Because the interface follows the OpenAI schema, existing clients and SDKs generally work after you change the model id and base URL. The supported parameters are include_reasoning, max_tokens, reasoning, temperature, tool_choice and tools. If you are building an agent, tools and tool_choice are the relevant ones, while reasoning and include_reasoning control how thinking is handled and max_tokens and temperature bound and shape the output.

There is no image or audio input here, so if your task needs multimodal input, another model will fit better. If you only need short, simple text generation with no tool calls, a smaller setup may be enough. Within its stated scope, Laguna XS 2.1 is aimed at coding agent workloads that need a long context and a generous output budget.