Laguna XS 2.1
Neural Network
Laguna XS 2.1 is a Poolside agentic model for code: generation, refactoring, and debugging via a unified BotHub API.
Max answer length
(in tokens)
Context size
(in tokens)
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Laguna XS 2.1?
Laguna XS 2.1 is a Poolside agentic model for coding, evolving the line after Laguna XS.2. It uses a 33B-A3B architecture: only a portion of parameters are active per request, balancing code analysis quality with response speed. It's a tool for engineering tasks—understanding project context, making precise edits, and executing multi-step scenarios. In BotHub, the model is accessible from Russia without a VPN or foreign card, and you pay in rubles only for what you use: tokens are deducted for actual requests and do not expire. With over 250 neural networks in the same window, you can compare responses and switch models without rewriting your integration, as the unified OpenAI-compatible API remains the same. Chats are encrypted via AES-GCM and not saved. For teams, we offer contracts, invoices, EDI, an admin panel, and limits. Use Laguna XS 2.1 when you need to write or improve code, understand a third-party repository, debug, prepare refactoring with tests, or integrate the model into your agent via API.Frequently asked questions about Laguna XS 2.1
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is a Poolside text model: it accepts text, documents, and images, reasons before answering, and calls external functions. A context of up to 262,144 tokens allows loading large files and code fragments in their entirety, making the model suitable for analysis, refactoring, debugging, analytics, and technical questions.
The model generates up to 32,768 tokens per response—enough for a voluminous document, detailed analysis, or a large code file. If the task is larger, break it into parts: the 262,144-token context allows you to keep the entire conversation and source code at hand.
Yes, reasoning before answering is confirmed for laguna-xs-2.1: the model first breaks down the task step-by-step, then formulates the result. This is noticeable in logic, mathematics, and code analysis, where it is important not to guess the answer, but to arrive at it sequentially.
Yes, the model supports function calling and structured JSON output—you can describe a schema and get a predictable response for parsing. In BotHub, it is available via a unified OpenAI-compatible API, so you can switch to another model without rewriting your integration.
Images and documents—yes, they are declared as supported input formats: you can send a screenshot of an error, a diagram, or a text document and ask questions about the content. There is no mention of audio or video in our data, so it is better to choose a model with explicit support for such files.
We do not have separate measurements for Russian, so we will not promise accuracy—the most honest way to check is to formulate your typical task and compare the answers. In BotHub, this is convenient: models are gathered in one window, access from Russia is available without a VPN or foreign card, and payment is in rubles.