Laguna XS.2
Poolside's first open model: a 33B coding agent model with 3B active, Apache 2.0. Replaced by XS 2.1 but still heavily downloaded.
In short. Laguna XS.2 is the first open model from Poolside, a small coding model with 33 billion parameters, of which 3 billion work on each token. It came out in April 2026 under Apache 2.0 and the model card says it runs on a Mac with 36 GB of memory. Poolside has since published an upgraded version, Laguna XS 2.1, but this one is still downloaded more than any other Laguna model.
Summary
Poolside announced Laguna XS.2 on 28 April 2026 as its first open-weights release, under Apache 2.0. Its technical report describes a 33.4B-parameter mixture of experts with 3B active parameters, trained from scratch, with only five weeks from the start of training to release. It has 256 experts plus one shared expert, 40 layers of which 10 use global attention and 30 use sliding-window attention over 512 tokens, an FP8 KV cache, a 256K-token context and interleaved thinking between tool calls that can be switched per request. Poolside later published Laguna XS 2.1 as an upgrade with the same size, a 5.4-point gain on SWE-bench Multilingual and a different license, so we list XS.2 as a legacy model.
In Poolside's own table, Laguna XS.2 scores 69.9% on SWE-bench Verified, 57.7% on SWE-bench Multilingual, 46.3% on SWE-Bench Pro and 35.7% on Terminal-Bench 2.0. In the same table Devstral Small 2 (24B dense) scores 68.0%, 55.7% and 22.5% on the three it reports, Qwen3.5-35B-A3B scores 69.2%, 60.3%, 44.6% and 40.5%, and Qwen3.6-35B-A3B scores 73.4%, 67.2%, 49.5% and 51.5%, so XS.2 trails the Qwen3.6 model on every column. Poolside took the highest published scores it could find for the comparison models, and the Gemma 4 31B scores in its table were reported by the Qwen team.
The BF16 weights are 66.9 GB, with FP8 at 35.3 GB, NVFP4 at 21.5 GB and INT4 at 23.9 GB. The card says the model is compact enough for a Mac with 36 GB of RAM and links an Ollama build with MLX support, and it also runs in vLLM, Transformers and TRT-LLM. In the 30 days to 11 November 2026, Hugging Face counted about 808,500 downloads of the main repository, more than any other Poolside repository, which shows how many deployments have not moved to XS 2.1. The Apache 2.0 license of XS.2 is also the more widely recognized one, since XS 2.1 uses OpenMDW-1.1.
Good for
- Existing local coding agents that already run XS.2 and do not need to move yet
- A small mixture-of-experts coding model under Apache 2.0 for a 36 GB Mac or a single GPU
- Teams whose compliance process accepts Apache 2.0 but has not reviewed OpenMDW-1.1
- Researchers who want to compare XS.2 with XS 2.1
Watch out for
- It is a legacy model, and Poolside's XS 2.1 improves on it with the same size, especially on SWE-bench Multilingual
- All scores are Poolside's own, and Qwen3.6-35B-A3B scores higher on all four benchmarks in its table
- The download count is high, but it may include automated pulls and existing deployments
- The card asks that you use the model consistently with Poolside's Acceptable Use Policy, which is separate from Apache 2.0
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the Laguna-XS.2 (BF16) variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| Laguna-XS.2 (BF16) BF16 | 33.4B 3B active | 77 GB | 38 GB | 23 GB | Gaming GPU |
| Laguna-XS.2-FP8 FP8 | 33.4B 3B active | 77 GB | 38 GB | 23 GB | Gaming GPU |
| Laguna-XS.2-NVFP4 NVFP4 | 33.4B 3B active | 77 GB | 38 GB | 24 GB | Gaming GPU |
| Laguna-XS.2-INT4 INT4 | 33.4B 3B active | 77 GB | 38 GB | 26 GB | Gaming GPU |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.
Frequently asked questions
- What is Laguna XS.2?
- Laguna XS.2 is the first open model from Poolside, a small coding model with 33 billion parameters, of which 3 billion work on each token. It came out in April 2026 under Apache 2.0 and the model card says it runs on a Mac with 36 GB of memory. Poolside has since published an upgraded version, Laguna XS 2.1, but this one is still downloaded more than any other Laguna model.
- Can I use Laguna XS.2 commercially?
- Yes. Laguna XS.2 is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is Laguna XS.2 open source?
- Laguna XS.2 is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
- Can I run Laguna XS.2 on my own computer?
- Yes, on a good graphics card. The smallest variant, Laguna-XS.2 (BF16), needs about 23 GB of memory at 4-bit precision, which suits a 24 to 32 GB gaming GPU or a Mac with 32 GB or more.
- When was Laguna XS.2 released?
- Laguna XS.2 was released by Poolside on April 28, 2026.
- How long a context window does Laguna XS.2 have?
- Laguna XS.2 supports a context window of up to 256K tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Introducing Laguna XS.2 and Laguna M.1 (Poolside, 28 April 2026) poolside.ai
- Laguna-XS.2 model card (Hugging Face) huggingface.co
- Laguna M.1 and XS.2 technical report (Poolside) poolside.ai
- Laguna-XS-2.1 model card, with the XS.2 comparison (Hugging Face) huggingface.co
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Laguna XS 2.1 Poolside | 33.4B 3B active | 256K | OpenMDW-1.1 (Permissive) | 22 GB Gaming GPU | |
| Laguna S 2.1 Poolside | 118B 8B active | 1M | OpenMDW-1.1 (Permissive) | 81 GB Workstation | |
| Laguna M.1 Poolside | 226B 23.4B active | 256K | Apache 2.0 (Permissive) | 153 GB Big-memory desktop | |
| Qwen3.6 Alibaba Qwen | 27.8B – 36B 3B active | 256K | Apache 2.0 (Permissive) | 19 GB Gaming GPU |