Laguna M.1-NVFP4 is a 225B total parameter Mixture-of-Experts model with 23B activated parameters per token designed for agentic coding and long-horizon work. This is the NVFP4-quantized variant of Laguna M.1.
[!NOTE]
This is the NVFP4 variant. The BF16 and FP8 variants are also available on Hugging Face.
Highlights
Large sparse MoE for agentic coding: Laguna M.1 is a 70-layer MoE transformer with 225B total parameters and 23B activated parameters per token
High-capacity expert routing: After 3 dense SwiGLU layers, Laguna M.1 uses 67 sparse MoE layers with 256 experts, top-k=16 routing and auxiliary-loss-free load balancing
Global attention architecture: Laguna M.1 uses global attention across all layers with 64 Q-heads, 8 KV-heads and softplus attention output gating
Native reasoning support: Interleaved thinking between tool calls with support for enabling and disabling thinking per-request
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: 225B total with 23B activated per token
Optimizer: Muon
Layers: 70 layers with global attention
Experts: 256 experts with 1 shared expert; top-k=16 routing
Dense layers: first 3 layers are dense SwiGLU; remaining 67 layers are sparse MoE
Attention: 64 Q-heads, 8 KV-heads, head dimension 128, with softplus attention output gating
Positional encoding: RoPE with YaRN
Modality: text-to-text
Context window: 262,144 tokens
Reasoning support: interleaved thinking with preserved thinking
Quantization: NVFP4 (weights), detected automatically from quantization_config
Benchmark results
benchmarks
Model
Parameters
SWE-bench Verified
SWE-bench Multilingual
SWE-bench Pro (Public Dataset)
Terminal-Bench 2.0
Laguna M.1 (BF16)
225B-A23B
74.6%
63.1%
49.2%
45.8%
Devstral 2
123B dense
72.2%
61.3%
-
32.6%
GLM-4.7
355B-A32B
73.8%
66.7%
-
41.0%
DeepSeek-V4 Flash
284B-A13B
79.0%
73.3%
52.6%
56.9%
Qwen3.5-397B-A17B
397B-A17B
76.2%
69.3%
50.9%
52.5%
Claude Sonnet 4.6
-
79.6%
-
-
59.1%
Scores shown are for the BF16 reference model; see the main Laguna M.1 model card for full benchmarking methodology. We used the highest publicly-referenced scores for all comparison models across each benchmark.
Usage
Laguna M.1 has upstream support in vLLM, SGLang, and TRT-LLM thanks to the support of the team at NVIDIA.
The full vLLM recipe is on the main Laguna M.1 model card. Quantization is detected automatically from quantization_config in this checkpoint, so the same command works with poolside/Laguna-M.1-NVFP4 substituted for the model ID. No extra flags required.
Laguna is supported in TensorRT-LLM thanks to the team at NVIDIA (NVIDIA/TensorRT-LLM#13559, with partial-RoPE fusion in #15110). The full recipe is on the main Laguna M.1 model card. Quantization is detected automatically from quantization_config in this checkpoint, so no extra flags are required.
Controlling reasoning
Laguna M.1 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. See the main Laguna M.1 model card for streaming, tool-call, and preserved-thinking examples.
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.
Laguna M.1 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 M.1 is subject to the Apache 2.0 License, and should be used consistently with Poolside's Acceptable Use Policy. We advise against circumventing Laguna M.1 safety guardrails without implementing substantially equivalent mitigations appropriate for your use case.
Please report security vulnerabilities or safety concerns to security@poolside.ai.
