**LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond**  
09.24.2026Models

### Related Links
- [09.23.2026News: Liquid AI, in Collaboration with Qualcomm Technologies, Brings Personal AI Context to Devices Powered by Snapdragon](/content/blog/liquid-context-qualcomm/index.html)  
- [09.17.2026Engineering: Setting the Frontier of Aging Biology with Liquid Foundation Models](/content/blog/longevitybench/index.html)  
- [09.17.2026Publications: An open benchmark and language models for AI in aging biology](https://www.cell.com/cell/fulltext/S0092-8674(26)00999-2)  
- [08.24.2026News: Introducing Pipette: A benchmarking suite for on-device intelligence](/content/blog/pipette-on-device-ai-benchmarking-by-liquid-ai/index.html)  
- [08.20.2026Models: LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook](/content/blog/lfm2.5-dspark/index.html)  
- [08.19.2026Models: LFM2.5 Q4_0: Quantization-Aware Distillation for Edge Deployment](/content/blog/qad/index.html)  
- [08.18.2026Engineering: Designing Loops for Production-Grade Work](/content/blog/agent-loops/index.html)  
- [08.12.2026Models: LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge](/content/blog/lfm2-5-vl-3b/index.html)  
- [08.05.2026News: MacPaw Partners with Liquid AI to Bring On-Device AI to Millions of Mac Users](/content/blog/macpaw-partners-liquid-ai-on-device-ai-mac-users/index.html)  
- [08.04.2026Models: LFM2.5-2.6B: Deploy Agents Everywhere](/content/blog/lfm2-5-2-6b/index.html)  
- [07.28.2026Models: LFM2.5-Encoders: Fast at Long Context, Even on CPU](/content/blog/lfm2-5-encoders/index.html)  
- [07.21.2026Engineering: Tokenizer Expansion: Upgrading a Model's Tokenizer in Place](/content/blog/tokenizer-expansion/index.html)  
- [07.16.2026Publications: In-Place Tokenizer Expansion for Pre-trained LLMs](https://arxiv.org/abs/2607.15232)  
- [07.07.2026Engineering: Reducing Doom Loops with Final Token Preference Optimization](/content/blog/antidoom/index.html)  
- [06.30.2026Engineering: IFStruct: Measuring structured-output compliance](/content/blog/ifstruct-v1.0)  
- [06.25.2026Models: LFM2.5-230M: Built to Run Anywhere](/content/blog/lfm2-5-230m/index.html)  
- [06.18.2026Models: LFM2.5 Retrievers: Bi-directional LFMs for Fast Multilingual Search](/content/blog/lfm2-5-retrievers/index.html)  
- [06.16.2026Publications: Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers](https://arxiv.org/abs/2606.18206)  
- [05.31.2026Publications: Neural Low-Discrepancy Sequences](https://arxiv.org/abs/2510.03745)  
- [05.29.2026Publications: SwitchCraft: A Programmatic Framework for Designing State-Switching Proteins](https://arxiv.org/abs/2605.31236)  
- [05.28.2026Models: LFM2.5-8B-A1B: An Even Better On-Device Mixture of Experts](/content/blog/lfm2-5-8b-a1b/index.html)  
- [05.20.2026Publications: MeMo: Model as a Memory](https://arxiv.org/abs/2605.15156)  
- [04.23.2026Press: Liquid AI and Mercedes-Benz partner to scale embedded in-car intelligence](/content/press/liquid-ai-and-mercedes-benz-partner-to-scale-embedded-in-car-intelligence/index.html)  
- [04.22.2026Publications: Preconditioned DeltaNet: Curvature-aware Sequence Modeling for Linear Recurrences](https://arxiv.org/abs/2604.21100)  
- [04.08.2026Models: LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud](/content/blog/lfm2-5-vl-450m/index.html)  
- [03.31.2026Models: LFM2.5-350M: No Size Left Behind](/content/blog/lfm2-5-350m-no-size-left-behind/index.html)  
- [03.26.2026Publications: The Key to State Reduction in Linear Attention: A Rank-based Perspective](https://openreview.net/forum?id=GfjoMXfaXq)  
- [03.16.2026External article: Why Bigger Isn’t Always Better in AI](https://www.bloomberg.com/opinion/articles/2026-03-16/why-bigger-isn-t-always-better-in-ai)  
- [03.05.2026Models: No Cloud, No Waiting: Tool-Calling Agents on Consumer Hardware with LFM2-24B-A2B](/content/blog/no-cloud-tool-calling-agents-consumer-hardware-lfm2-24b-a2b/index.html)  
- [03.03.2026Press: Liquid AI and Insilico Medicine Announce Strategic Partnership Delivering Lightweight Scientific Foundation Models for Drug Discovery](/content/press/liquid-ai-insilico-medicine-partnership/index.html)