Poolside: Laguna S 2.1

Poolside: Laguna S 2.1 is a text model from Poolside, listed on api14 under the model id poolside/laguna-s-2.1. Poolside describes it as the latest coding agent model in its Laguna line. It is a 118B total parameter model with 8B active parameters, which means the architecture activates a smaller subset of weights per token rather than the full parameter count. The provider release date is 2026-07-21.

The model works with text input and produces text output, so it fits text generation and agent style workloads rather than image, audio or other modalities. It has a context length of 1,048,576 tokens, which is large enough to hold very long inputs such as whole repositories, long logs or extended conversation histories in a single request. The maximum output is 131,072 tokens, so responses can run well beyond a short answer when the task calls for it. Poolside reports a score of 70.2% on Terminal-Bench 2.1, a benchmark oriented around terminal and agent tasks.

Those capabilities point to a few practical uses. Coding agent workflows are the clearest fit, given the model's positioning and its Terminal-Bench result: running multi step tasks, reading and editing files, and working through command line style problems. The long context supports repository level analysis, where a developer needs the model to reason over many files at once, and it also suits long document or log review where the relevant details are spread across a large input. Because the model supports tools and reasoning, it can be used in setups where the model decides when to call a function and where the reasoning trace is exposed to the caller.

You call it through api14's OpenAI compatible API using the model id poolside/laguna-s-2.1. The supported request parameters are include_reasoning, max_tokens, reasoning, temperature, tool_choice and tools. In practice that means you can cap the response length with max_tokens, control sampling with temperature, pass tool definitions through tools and steer selection with tool_choice, and manage reasoning output with the reasoning and include_reasoning parameters.

There is no information here about pricing, rate limits or latency, so those need to be checked on the live price table and in your own testing. If your task is not text based, or does not involve coding or agent style work, another model may fit better. For straightforward short text tasks, the very large context and output limits may be more than you need.