Laguna S 2.1

Laguna S 2.1 sits inside the broader laguna family and is positioned as a coding-oriented large language model. Independent coverage from the Latent Space AINews roundup on July 23, 2026 frames the release as a competitive open-weights entry, with a community-discussed narrative that it undercuts the cost of Deepseek v4 Flash while approaching or exceeding Deepseek v4 Pro on benchmarks, reigniting discussion around distillation across open coding models. The piece situates the release alongside other Western "neolab" efforts that aim to match larger Chinese model equivalents with smaller, more efficient footprints.

The intended use case for Laguna S 2.1 is software engineering assistance and general text reasoning, building on a lineage of open coding models that prioritise accessible weights and developer workflow integration. Practically, the model is aimed at teams who want an open-weights alternative for code generation, refactoring, and reasoning-heavy assistant tasks without depending on closed-API frontier systems. Its placement in a community discussion that emphasises both price and benchmark performance against established coding models suggests it is best suited for cost-sensitive deployments where open weights and reproducibility matter more than absolute top-tier benchmark dominance.

Vultrlaguna-s-2.1laguna

Open provider page Provider docs

Quick Info

Provider: Vultr
Model key: laguna-s-2.1
Release date: Jul 21, 2026
Last updated: Jul 21, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.09
Output token cost: $0.18

Limits

Output tokens: 32,768 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare Laguna S 2.1 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Laguna S 2.1

Vultr

Coverage

Poolside released Laguna S 2.1, a 118-billion-parameter open-weight foundation model purpose-built for agentic coding. On Terminal-Bench 2.1 and SWE-Bench Pro, it matches or exceeds models several times its size, including DeepSeek-V4-Flash, NVIDIA Nemotron 3 Ultra, and Thinking Machines' Inkling. It is small enough to run on a single NVIDIA DGX Spark. The weights are available on Hugging Face under the OpenMDW-1.1 license, and Poolside also published reproducible agentic-coding evaluations run using its agent harness called "pool." Laguna S 2.1 is positioned as the first scale-up from Poolside's earlier XS release, aimed at self-hosted, sovereignty-friendly open-weight deployment.

Vultr

Coverage

Laguna S 2.1 is a sparse Mixture-of-Experts transformer with 118 billion total / 8 billion active parameters, supporting both thinking and no-thinking modes with a 1,048,576-token context window. The architecture uses 48 layers of 256 routed experts plus one shared expert, grouped-query attention with 8 KV heads, and interleaved global and 512-token sliding-window attention layers. Pre-training began May 22, 2026 and the model launched July 21, under nine weeks end-to-end. Weights are published on Hugging Face in BF16, FP8, INT4 and NVFP4 formats with official GGUF and MLX conversions under the OpenMDW-1.1 license, and run on vLLM, SGLang, Transformers, TRT-LLM, and llama.cpp. Poolside reports Terminal-Bench 2.1 at 70.2% and SWE-bench Pro at 59.4%.

Vultr

CoverageRelease Notes

Poolside announced Laguna S 2.1, a 118-billion-parameter open-weight foundation model for agentic coding, available today under an OpenMDW-1.1 license. On Terminal-Bench 2.1 and SWE-Bench Pro, it matches or exceeds much larger models including DeepSeek-V4-Flash, NVIDIA Nemotron 3 Ultra, and Thinking Machines' Inkling, and runs on a single NVIDIA DGX Spark. Weights are available on Hugging Face, with reproducible evaluations run using Poolside's agent harness "pool." The release is framed as a Western open-weight option competing on cost, control, and self-hosting for sovereignty, positioned as the first scale-up from Poolside's earlier XS model.