README.md

---
library_name: vllm
inference: false
extra_gated_description: >-
  To learn more about how we process your personal data, please read our <a
  href="https://poolside.ai/legal/privacy">Privacy Policy</a>.
tags:
- laguna-xs.2
license: apache-2.0
pipeline_tag: text-generation
---

<p align="center">
  <img alt="poolside-banner" src="https://poolside.ai/assets/laguna/laguna-xs2-banner.svg" width="800px">
</p>

<p align="center">
  <a href="https://shimmer.poolside.ai"><strong>Try Laguna XS.2 in Shimmer</strong></a> ·
  <a href="https://platform.poolside.ai"><strong>Get an API key</strong></a> ·
  <a href="https://poolside.ai/blog/laguna-a-deeper-dive"><strong>Release blog post</strong></a>
</p>

<br>

# Laguna XS.2-INT4
Laguna XS.2-INT4 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine. It uses Sliding Window Attention with per-head gating in 30 out of 40 layers for fast inference and low KV cache requirements.

> [!NOTE]
> This is the INT4 variant with an FP8-quantized KV cache. The [BF16](https://huggingface.co/poolside/Laguna-XS.2), [FP8](https://huggingface.co/poolside/Laguna-XS.2-FP8) and [NVFP4](https://huggingface.co/poolside/Laguna-XS.2-NVFP4) variants are also available on Hugging Face.

## Highlights
- **Mixed SWA and global attention layout**: Laguna XS.2 uses sigmoid gating with per-layer rotary scales, enabling mixed SWA (Sliding Window Attention) and global attention layers in a 3:1 ratio (across 40 total layers)
- **KV cache in FP8**: KV cache quantized to FP8, reducing memory per token
- **Native reasoning support**: Interleaved thinking between tool calls with support for enabling and disabling thinking per-request
- **Local-ready**: At 33B total parameters and 3B activated, Laguna XS.2 is compact enough to run on a Mac with 36 GB of RAM. [Available on Ollama](https://ollama.com/library/laguna-xs.2)
- **Apache 2.0 license**: Use and modify freely for commercial and non-commercial purposes

---

## Model overview

- Training: pre-training, post-training and reinforcement learning stages
- Number of parameters: 33B total with 3B activated per token
- Optimizer: Muon
- Layers: 40 layers (10 layers with global attention, 30 layers with sliding window attention)
- Experts: 256 experts with 1 shared expert
- Sliding Window: 512 tokens
- Modality: text-to-text
- Context window: 131,072 tokens
- Reasoning support: interleaved thinking with preserved thinking

## Benchmark results

<p align="center">
  <img alt="benchmarks" src="https://poolside.ai/assets/laguna/laguna-xs2-chart.svg" width="800px">
</p>

| Model                     | Size (total params.) | SWE-bench Verified | SWE-bench Multilingual | SWE-bench Pro (Public Dataset) | Terminal-Bench 2.0 |
|---------------------------|----------------------|--------------------|------------------------|--------------------------------|--------------------|
| **Laguna XS.2 (BF16)**    | 33B                  | 68.2%              | 62.4%                  | 44.5%                          | 30.1%              |
| Devstral Small 2          | 24B dense            | 68.0%              | 55.7%                  | -                              | 22.5%              |
| Gemma 4 31B IT            | 31B dense            | 52.0%              | 51.7%                  | 35.7%                          | 42.9%              |
| Qwen3.5-35B-A3B           | 35B                  | 69.2%              | 60.3%                  | 44.6%                          | 40.5%              |
| Qwen3.6-35B-A3B           | 35B                  | 73.4%              | 67.2%                  | 49.5%                          | 51.5%              |
| Claude Haiku 4.5          | -                    | 73.3%              | -                      | 39.5%                          | 29.8%              |
| GPT-5.4 Nano              | -                    | -                  | -                      | 52.4%                          | 46.3%              |

*We used the highest publicly-referenced scores for all comparison models across each benchmark. In almost all cases these were official scores published in release blog posts or equivalent, with the exception of Gemma 4 31B IT where the highest published scores were [reported by the Qwen team](https://qwen.ai/blog?id=qwen3.6-35b-a3b) and Claude Haiku 4.5 where the highest published (verified) scores for SWE-bench Pro and Terminal-Bench 2.0 are from their respective official leaderboards.*

<details>
<summary>Expand for benchmarking methodology</summary>

All benchmarking for Laguna XS.2 was completed using the Laude Institute’s Harbor Framework with our [agent harness](https://github.com/poolsideai/pool), using a maximum of 500 steps and sandboxed execution using 8 GB RAM/2 CPUs (with the exception of Terminal-Bench 2.0; see below). The same sampling parameters were used for all benchmarking: temperature=0.7 and top_k=20.  Some base task images and verifiers were patched to fix infrastructure reliability issues inherent in task setup, such as rate limits on third-party dependencies in external registries used by the verifier. More details outlining these updates and other findings will follow in a future technical blog post.

- SWE-bench Verified: mean pass@1 averaged over 4 runs.
- SWE-bench Multilingual: mean pass@1 averaged over 7 runs.
- SWE-bench Pro: mean pass@1 averaged over 3 runs.
- Terminal-Bench 2.0: mean pass@1 averaged over 5 runs. 48GB RAM/32 CPUs.

</details>

## Usage

Laguna XS.2-INT4 has launch-day support in vLLM and Transformers.

The fastest way to get started is with our API, directly or using OpenRouter.

> [!NOTE]
> For complete usage instructions, see the main [Laguna XS.2 model card](https://huggingface.co/poolside/Laguna-XS.2).

### Local deployment

Laguna XS.2-INT4 is supported in vLLM and Transformers. Use Laguna-XS.2 with Ollama (with MLX support) and the mlx-lm framework for the best experience on your local machine.

#### vLLM

The full vLLM recipe is on the main [Laguna XS.2 model card](https://huggingface.co/poolside/Laguna-XS.2). Quantization is detected automatically from `quantization_config` in this checkpoint, so the same command works with `poolside/Laguna-XS.2-INT4` substituted for the model ID. No extra flags required.

> [!NOTE]
> Please note that, during testing, we discovered that models with FP8-quantised KV caches can produce scrambled output when deployed on non-Hopper GPUs. We are actively investigating this issue with the vLLM team, but in the meantime, you can circumvent this issue by explicitly disabling FP8 KV cache (Laguna XS.2 has 40 layers, so list every layer in `--kv-cache-dtype-skip-layers`):
>
> ```shell
> vllm serve poolside/Laguna-XS.2-INT4 \
>     --kv-cache-dtype-skip-layers $(seq 0 39) \
>     --max-model-len 131072 \
>     --reasoning-parser poolside_v1 \
>     --tool-call-parser poolside_v1
> ```
>
> The [BF16 checkpoint](https://huggingface.co/poolside/Laguna-XS.2) is unaffected as it does not declare an FP8 KV cache.

#### Transformers

The full Transformers recipe is on the main [Laguna XS.2 model card](https://huggingface.co/poolside/Laguna-XS.2). Substitute `poolside/Laguna-XS.2-INT4` for the model ID; quantization is detected automatically from `quantization_config`.

#### Ollama

Visit [Ollama's model library](https://ollama.com/library/laguna-xs.2) to pull to your local machine.

## Controlling reasoning

Laguna XS.2 has native reasoning support and is designed to work best with *preserved thinking*, where `reasoning` content from prior assistant messages is preserved in the message history. This model will generally reason before calling tools and between tool calls.

<details>
<summary>Expand for example</summary>

```python
import json
from openai import OpenAI

client = OpenAI(
  base_url="https://inference.poolside.ai/v1",
  api_key="...",
)

model = "poolside/laguna-xs.2"

tools = [{"type": "function", "function": {
  "name": "shell",
  "description": "Execute a bash command and return the output.",
  "parameters": {"type": "object", "properties": {"cmd": {"type": "string"}}, "required": ["cmd"]},
}}]

messages = [
  {"role": "system", "content": "You are a coding agent with access to a shell tool."},
  {"role": "user", "content": "Run uname -a"},
]

# Thinking is enabled by default when the server sets --default-chat-template-kwargs {"enable_thinking": True}
# When using the Poolside API (https://inference.poolside.ai/v1), this flag is set by default
response = client.chat.completions.create(
  model=model,
  messages=messages,
  tools=tools,
  stream=True,
)

reasoning, content, tool_calls = "", "", []
for chunk in response:
  delta = chunk.choices[0].delta
  if hasattr(delta, "reasoning_content") and delta.reasoning_content:
    reasoning += delta.reasoning_content
  if hasattr(delta, "content") and delta.content:
    content += delta.content
  if hasattr(delta, "tool_calls") and delta.tool_calls:
    for tc in delta.tool_calls:
      if tc.index >= len(tool_calls):
        tool_calls.append({"id": tc.id, "function": {"name": "", "arguments": ""}})
      if tc.function.name:
        tool_calls[tc.index]["function"]["name"] = tc.function.name
      if tc.function.arguments:
        tool_calls[tc.index]["function"]["arguments"] += tc.function.arguments

print(f"Reasoning: {reasoning}\nContent: {content}\nTool calls: {tool_calls}\n")

# Return reasoning in the next request for best performance
messages.append({
  "role": "assistant",
  "content": content,
  "reasoning_content": reasoning,
  "tool_calls": [{"id": tc["id"], "type": "function", "function": tc["function"]} for tc in tool_calls]
})

messages.append({
  "role": "tool",
  "tool_call_id": tool_calls[0]["id"],
  "content": json.dumps({"stdout": "Darwin arm64", "exit_code": "0"})
})

response = client.chat.completions.create(
  model=model,
  messages=messages,
  tools=tools,
  stream=True,
)

reasoning, content = "", ""
for chunk in response:
  delta = chunk.choices[0].delta
  if hasattr(delta, "reasoning_content") and delta.reasoning_content:
    reasoning += delta.reasoning_content
  if hasattr(delta, "content") and delta.content:
    content += delta.content

print(f"Reasoning: {reasoning}\nContent: {content}")
```

</details>

### Disabling reasoning

You can disable thinking by setting `enable_thinking` to `False` in a request or by not providing `--default-chat-template-kwargs {"enable_thinking": True}` or equivalent when starting the server.

<details>
<summary>Expand for example</summary>

```python
from openai import OpenAI
client = OpenAI()

completion = client.chat.completions.create(
  model="poolside/laguna-xs.2",
  messages=[
    {"role": "user", "content": "Write a retry wrapper with exponential backoff."}
  ],
  extra_body={
    "chat_template_kwargs": { "enable_thinking": False },
  },
  stream=True
)

for chunk in completion:
    print(chunk.choices[0].delta)
```

</details>

For agentic coding use cases, we recommend enabling thinking and preserving reasoning in message history as outlined in the [Controlling reasoning] section.

## License

This model is licensed under the [Apache 2.0 License](https://huggingface.co/poolside/Laguna-XS.2-INT4/blob/main/LICENSE.md).

## Intended and Responsible Use 

Laguna XS.2-INT4 is designed for software engineering and agentic coding use cases, and you are responsible for confirming that it is appropriate for your intended application. Laguna XS.2-INT4 is subject to the [Apache 2.0 License](https://huggingface.co/poolside/Laguna-XS.2-INT4/blob/main/LICENSE.md), and should be used consistently with Poolside's [Acceptable Use Policy](https://poolside.ai/legal/acceptable-use-policy). We advise against circumventing Laguna XS.2-INT4 safety guardrails without implementing substantially equivalent mitigations appropriate for your use case.

Please report security vulnerabilities or safety concerns to [security@poolside.ai](mailto:security@poolside.ai).