Laguna XS 2.1

Poolside

Laguna XS 2.1 is Poolside's open-weight Mixture-of-Experts coding model with 33B total parameters and 3B activated per token (256 experts + 1 shared; 40 layers). It targets agentic coding and long-horizon local work, with interleaved thinking, a 262,144-token context window, and text-to-text only modalities. Available as open weights on Hugging Face and via Poolside's API and OpenRouter (poolside/laguna-xs-2.1). It is served by poolside with a 256K-token context window at $0.1 / $0.2 per 1M input/output tokens.

Key Specifications

Parameters
33.0B
Context
262.1K
Release Date
July 2, 2026
Average Score
54.8%

Timeline

Key dates in the model's history
Announcement
July 2, 2026
Last Update
September 12, 2026
Today
September 19, 2026

Technical Specifications

Parameters
33.0B
Training Tokens
-
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.10
Output (per 1M tokens)
$0.20
Max Input Tokens
262.1K
Max Output Tokens
-
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests
SWE-Bench Verified
Harbor + Poolside agent harness; thinking on; 256K ctx; temp=1.0, top_k=20, top_p=1; mean pass@1 over 4 attemptsSelf-reported
70.9%

Other Tests

Specialized benchmarks
SWE-bench Multilingual
Harbor + Poolside agent harness; thinking on; 256K ctx; temp=1.0, top_k=20, top_p=1; mean pass@1 over 4 attemptsSelf-reported
63.1%
SWE-Bench Pro
Harbor + Poolside agent harness; thinking on; 256K ctx; temp=1.0, top_k=20, top_p=1; mean pass@1 over 2 attempts; public setSelf-reported
47.6%
Terminal-Bench 2.0
Harbor + Poolside agent harness; thinking on; 256K ctx; temp=1.0, top_k=20, top_p=1; mean pass@1 over 5 attempts; 48 GB RAM/32 CPUsSelf-reported
37.5%

License & Metadata

License
openmdw
Announcement Date
July 2, 2026
Last Updated
September 12, 2026

Compare Laguna XS 2.1

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.