Poolside releases Laguna XS.2, free fp8-quantized coding agent with 128K context

Poolside has released Laguna XS.2, the second-generation model in its XS size class designed for agentic coding workflows. The model is available free on OpenRouter as of April 28, 2025.

Technical specifications

Laguna XS.2 offers a 131,072-token context window (128K) with up to 8K output tokens. The model is quantized to fp8 precision, optimizing for speed and cost efficiency in production environments.

According to Poolside, the model combines tool calling and reasoning capabilities within a compact footprint. The company describes it as part of their "efficient coding agent series."

Pricing and availability

The model is available at zero cost through OpenRouter:

• Input: $0 per million tokens
• Output: $0 per million tokens

OpenRouter routes requests across multiple providers with automatic fallbacks for uptime optimization.

Reasoning capabilities

Laguna XS.2 supports OpenRouter's reasoning parameter, allowing developers to access step-by-step thinking processes through the reasoning_details array in API responses. The model can preserve reasoning context across conversation turns when the complete reasoning_details are passed back in subsequent requests.

What this means

Poolside is positioning itself in the increasingly competitive coding agent market with a free, quantized model that prioritizes deployment efficiency over raw capability. The fp8 quantization represents a pragmatic trade-off—reduced precision for faster inference—targeting production workflows where cost and latency matter more than maximum accuracy. At 128K context, Laguna XS.2 can handle substantial codebases, though it remains to be seen how the XS size class compares to larger coding models like Claude 3.5 Sonnet or GPT-4 on complex refactoring tasks. The free tier may be a customer acquisition strategy, with Poolside likely planning premium tiers or enterprise offerings.