Poolside releases Laguna XS 2.1: 33B parameter MoE coding model with 262K context window
Poolside has released Laguna XS 2.1, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token and a 262,144-token context window. The model achieves 70.9% on SWE-bench Verified and 63.1% on SWE-bench Multilingual, representing a 5.4% improvement over its predecessor on multilingual coding tasks.
Laguna XS 2.1 at a glance
How Laguna XS 2.1 ranks now
| Benchmark | Score | Rank | Source |
|---|---|---|---|
| SWE-bench Verified | 70.9% | 57 of 80 | Hugging Face model card, Jul 2026 ↗ |
Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.
Poolside releases Laguna XS 2.1: 33B parameter MoE coding model with 262K context window
Poolside has released Laguna XS 2.1, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token and a 262,144-token context window. The model achieves 70.9% on SWE-bench Verified and 63.1% on SWE-bench Multilingual, representing a 5.4% improvement over its predecessor on multilingual coding tasks.
Model specifications
Laguna XS 2.1 uses a 256-expert architecture with 1 shared expert across 40 layers in a 3:1 ratio of sliding window attention (512-token window) to global attention layers. The model employs sigmoid gating with per-layer rotary scales and quantizes its KV cache to FP8 to reduce memory usage. According to Poolside, the model can run on a Mac with 36GB of RAM.
The model was trained using pre-training, post-training, and reinforcement learning stages with the Muon optimizer. It supports native reasoning with interleaved thinking between tool calls that can be enabled or disabled per request.
Benchmark performance
On SWE-bench Verified, Laguna XS 2.1 scored 70.9%, up from 69.9% in version XS.2. The model achieved 63.1% on SWE-bench Multilingual (versus 57.7% previously), 47.6% on SWE-Bench Pro, and 37.5% on Terminal-Bench 2.0.
For comparison, Qwen3.6-35B-A3B scored 73.4% on SWE-bench Verified and 67.2% on SWE-bench Multilingual, while Claude Haiku 4.5 achieved 73.3% on SWE-bench Verified. All benchmarks were run using the Harbor Framework with temperature=1.0, top_k=20, top_p=1, and thinking mode enabled at 256K context length.
Availability and deployment
Laguna XS 2.1 is available with free inference on OpenRouter for a limited time. The model has day-one support in vLLM (version 0.21.0+), SGLang, Transformers (v5.7.0+), Llama.cpp, and TensorRT-LLM (v1.3.0rc16+).
Poolside offers optional speculative decoding through DFlash, a 5-layer draft model that proposes up to 7 tokens per step with approximately 70% per-position acceptance on coding tasks, according to the company.
The model is released under the OpenMDW-1.1 license, which permits commercial and non-commercial use and modification. Poolside also provides a terminal-based coding agent called "pool" with Agent Client Protocol support for macOS and Linux.
What this means
Laguna XS 2.1 targets the growing space of local coding agents, offering competitive SWE-bench scores in a package small enough for consumer hardware. The 5.4% improvement on multilingual coding tasks suggests focused optimization for non-English programming scenarios. However, the model trails larger competitors like MAI-Code-1-Flash (137B parameters) which scores 71.6% on SWE-bench Verified and 65.5% on Multilingual, indicating that parameter efficiency comes with performance tradeoffs on certain benchmarks.
Related Articles
Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context
Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.
Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license
Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.
Paradigm Intelligence launches Phai-A1 language model and Phai-M1 video model, trained on Huawei Ascend
Paradigm Intelligence (范式智能, formerly 第四范式) has unveiled two in-house models: Phai-A1, a language model that runs on a single RTX 3090, and Phai-M1, a video model with synchronized audio. The company says both were trained on Huawei Ascend hardware, with open weights coming soon.
Tencent releases Hy4 flash: 220B MoE, 12B active parameters, 1M-token context
Tencent has released Hy4 flash, a multimodal Mixture-of-Experts model with 220B total parameters, 12B active per token, and a 1,048,576-token context window. Tencent calls it its most cost-effective Hy model to date and claims a per-task cost of 1/20 that of Hy4 on its internal WorkIndex benchmark.
Comments
Loading...