Poolside releases Laguna XS.2: 33B parameter MoE coding model with 131K context window
Poolside has released Laguna XS.2, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token, designed for agentic coding. The model features a 131,072-token context window, scores 68.2% on SWE-bench Verified, and is available under Apache 2.0 license with free API access.
Poolside releases Laguna XS.2: 33B parameter MoE coding model with 131K context window
Poolside has released Laguna XS.2, a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token, designed for agentic coding and long-horizon work on local machines.
Model specifications
Laguna XS.2 uses Sliding Window Attention with per-head gating in 30 of its 40 layers. The architecture includes:
- Total parameters: 33B with 3B activated per token
- Context window: 131,072 tokens
- Experts: 256 experts with 1 shared expert
- Architecture: 40 layers (10 global attention, 30 sliding window attention)
- Sliding window: 512 tokens
- Training: Muon optimizer, includes pre-training, post-training, and reinforcement learning stages
- License: Apache 2.0
The model uses FP8-quantized KV cache to reduce memory requirements and supports native reasoning with interleaved thinking between tool calls.
Benchmark performance
According to Poolside, Laguna XS.2 achieves the following scores:
- SWE-bench Verified: 68.2% (mean pass@1 over 4 runs)
- SWE-bench Multilingual: 62.4% (mean pass@1 over 7 runs)
- SWE-bench Pro: 44.5% (mean pass@1 over 3 runs)
- Terminal-Bench 2.0: 30.1% (mean pass@1 over 5 runs)
Poolside claims these scores place it competitively with Devstral Small 2 (24B dense, 68.0% on SWE-bench Verified) while using fewer activated parameters. Qwen3.6-35B-A3B leads the comparison group at 73.4% on SWE-bench Verified.
All benchmarking was completed using the Laude Institute's Harbor Framework with temperature=0.7 and top_k=20. The company notes some task images and verifiers were patched to fix infrastructure reliability issues.
Availability and deployment
Poolside is offering free API access to Laguna XS.2 and its larger 225B model, Laguna M.1, for a limited time. Pricing details have not been disclosed.
The model has launch-day support in:
- vLLM (pending PR merge)
- Transformers (support merged, shipping in release after v5.6.2)
- TRT-LLM (pending upstream PR)
- Ollama with MLX support
The company has released pool, a lightweight terminal-based coding agent that works with the model. Laguna XS.2 can run on a Mac with 36 GB of RAM according to Poolside.
What this means
Laguna XS.2 represents a push toward locally-runnable coding models with competitive benchmark performance. The 33B total parameter count with 3B activation makes it accessible for developers with high-end consumer hardware, while the Apache 2.0 license removes commercial restrictions. The 131K context window matches or exceeds many commercial models, potentially enabling longer coding sessions without context truncation. However, its 68.2% SWE-bench Verified score trails the latest Qwen and Claude models, suggesting it may be best suited for local development workflows where privacy and control outweigh absolute performance.
Related Articles
Tencent Open-Sources AuK, a 1.5B-Parameter Speech Generation and Editing Model
Tencent has open-sourced AuK, a 1.5B-parameter foundation model for speech generation and editing that handles TTS, content editing, and audio enhancement through natural-language instructions. The release includes a distilled AuK-Flash variant for 4-step fast inference, both under MIT license.
DeepSeek Launches V4.1 Flash: Low-Cost MoE Model Claims to Beat V4 Pro
DeepSeek has released V4.1 Flash, a sparse mixture-of-experts model priced at $0.30 per 1M input tokens and $1.20 per 1M output tokens with a 1 million token context window. DeepSeek claims the model exceeds the larger V4 Pro on performance, speed, and task completion time.
DeepSeek Releases V4.1-Flash: 552B MoE Model Cuts KV Cache to 890 Bytes Per Token
DeepSeek has released V4.1-Flash, a 552B-parameter multimodal Mixture-of-Experts model supporting 1M-token context and activating only 8B parameters during prefill. The model uses a new Causal Encoder-Decoder architecture and Compressed Sparse Attention 2 to cut global KV cache to 890 bytes per token, roughly a quarter of its predecessor.
AllSpark's Iris-mini and Iris-pro Top Open-Weight Search Agent Benchmarks
Chinese lab AllSpark has released Iris-mini and Iris-pro, two open-weight search agents built on Qwen3 models that claim the top spot among open-weight systems in their size classes on four research benchmarks. The release includes model weights, an agent harness, and evaluation code, with training pipelines to follow.
Comments
Loading...