PoolsideLaguna
Laguna S 2.1
Laguna S 2.1 has 118 billion total and about 8 billion active parameters. It accepts text and supports reasoning between tool calls.
What the release provides
The 48-layer architecture mixes global and sliding-window attention. Poolside supplies BF16, FP8, NVFP4, INT4 and GGUF packages plus a DFlash draft model for speculative decoding. The release uses OpenMDW 1.1.Poolside: Laguna S 2.1 model card
How to use it well
Practical guidance. Size memory for the full weights, context cache and concurrent requests. Sparse activation reduces computation per token; it does not remove inactive weights from memory.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable, Provider product |
| Model identifiers | poolside/Laguna-S-2.1 |
| Context window | 1,048,576 tokens Published model context window Poolside: Laguna S 2.1 model card |
| Thinking settings | Off or Max; intermediate reasoning-effort levels are not supportedPoolside: Model release notes |
| Supported deployment | Included in Poolside’s self-managed inference offeringPoolside: Supported models |
Related models
- Laguna XS 2.1
Poolside · Laguna
A smaller Laguna release for coding agents on local hardware.
Launch date not confirmedDownloadable · Provider product - Laguna S 2.1 base
Poolside · Laguna
A research-access base checkpoint underlying Laguna S 2.1.
Launch date not confirmedRestricted accessDownloadable