Laguna XS 2.1
Description
Laguna XS 2.1 is a Poolside 33B total parameter Mixture-of-Experts text generation model with 3B activated parameters per token, designed for agentic coding and long-horizon software engineering work on local machines.
This model is ready for commercial or non-commercial use.
Third-Party Community Consideration:
This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party's requirements for this application and use case; see link to Non-NVIDIA Laguna XS 2.1 Model Card
License and Terms of Use:
GOVERNING TERMS: The trial service is governed by the NVIDIA API Trial Terms of Service. Use of the model is governed by the OpenMDW License Agreement, version 1.1.
Deployment Geography:
Global
Use Case:
Use Case: Laguna XS 2.1 is intended for software engineering, agentic coding, terminal-style tasks, tool-use workflows, long-horizon coding work, and local text generation.
Release Date:
Build.NVIDIA.com: 07/15/2026 via link
Huggingface: 07/02/2026 via link
Reference(s):
References:
Model Architecture:
Architecture Type: Transformer
Network Architecture: Mixture-of-Experts
Total Parameters: 33B
Active Parameters: 3B
Vocabulary Size: 100,352
Input:
Input Types: Text
Input Formats: String
Input Parameters: One-Dimensional (1D)
Other Input Properties: Laguna XS 2.1 uses a chat template that supports optional thinking, tool calls, and preserved reasoning content.
Input Context Length (ISL): 262,144
Output:
Output Types: Text
Output Format: String
Output Parameters: One-Dimensional (1D)
Other Output Properties: The model can generate text with interleaved reasoning and tool-call content when supported by the serving stack.
Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.
Software Integration:
Runtime Engines:
- llama.cpp
- Ollama
- SGLang
- TensorRT-LLM
- Transformers
- vLLM
Supported Hardware:
- NVIDIA Blackwell: B200, B100, GB200
- NVIDIA Hopper: H100, H200
Preferred Operating Systems: Linux
The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.
Model Version(s)
Laguna XS 2.1 v2.1
Training, Testing, and Evaluation Datasets:
Training Dataset
Data Modality: Text
Text Training Data Size: Undisclosed
Training Data Collection: Automated
Training Labeling: Automated
Training Properties: Laguna XS 2.1 was developed through pre-training, post-training, and reinforcement learning stages. Specific training datasets are Undisclosed.
Testing Dataset
Testing Data Collection: Undisclosed
Testing Labeling: Undisclosed
Testing Properties: Undisclosed
Evaluation Dataset
Evaluation Benchmark Score: Laguna XS 2.1 reports 70.9% on SWE-bench Verified, 63.1% on SWE-bench Multilingual, 47.6% on SWE-Bench Pro, and 37.5% on Terminal-Bench 2.0.
Detailed Benchmark Comparison Table
| Model | Size (total params.) | SWE-bench Verified | SWE-bench Multilingual | SWE-Bench Pro (Public Dataset) | Terminal-Bench 2.0 |
|---|---|---|---|---|---|
| Laguna XS 2.1 | 33B | 70.9% | 63.1% | 47.6% | 37.5% |
| Laguna XS.2 | 33B | 69.9% | 57.7% | 46.3% | 35.7% |
| Qwen3.6-35B-A3B | 35B | 73.4% | 67.2% | 49.5% | 51.5% |
| North Mini Code | 30B | 67.6% | - | 40.2% | 36.0% |
| MAI-Code-1-Flash | 137B | 71.6% | 65.5% | 51.2% | 54.8% |
| gpt-oss-120B | 120B | - | - | 16.2% | 18.7% |
| Claude Haiku 4.5 | - | 73.3% | - | 39.5% | 29.8% |
| GPT-5.4 Nano | - | - | - | 52.4% | 46.3% |
Evaluation Data Collection: Hybrid: Automated, Manually-Collected
Evaluation Labeling: Hybrid: Automated, Manually-Labeled
Evaluation Properties: Evaluation benchmarks include SWE-bench Verified, SWE-bench Multilingual, SWE-Bench Pro, and Terminal-Bench 2.0. Evaluation Methodology Notes:
- All Laguna XS 2.1 benchmarking used Laude Institute's Harbor Framework with Poolside's agent harness, a maximum of 500 steps, sandboxed execution, temperature=1.0, top_k=20, top_p=1, thinking mode enabled, and a 256K-token context length.
- SWE-bench Verified and SWE-bench Multilingual report mean pass@1 averaged over 4 attempts per task.
- SWE-Bench Pro reports mean pass@1 averaged over 2 attempts per task.
- Terminal-Bench 2.0 reports mean pass@1 averaged over 5 attempts per task using 48 GB RAM and 32 CPUs; other tasks used 8 GB RAM and 2 CPUs.
- Base task images and verifiers were patched for infrastructure reliability issues, and a reward-hack judge review did not find significant reward hacking after joint judge review and manual review.
Inference
Acceleration Engine: vLLM
Test Hardware: NVIDIA Hopper (H100)
Ethical Considerations
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.
